Merge branch 'main' into worktree-dns-route-qtype-fallthrough

Co-authored-by: braginini <bangvalo@gmail.com>

.github/workflows/agent-network-e2e.yml

@@ -5,6 +5,13 @@ on:
   schedule:
     - cron: "0 3 * * *"
   workflow_dispatch:
+    inputs:
+      bedrock_model:
+        description: >-
+          Bedrock inference-profile id to drive the matrix with, exactly as
+          AWS issues it. Leave empty for the Sonnet 4.6 default.
+        required: false
+        default: ""
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -51,6 +58,9 @@ jobs:
           # token (and URL, for gateways) is unset, so partial coverage is fine.
           OPENAI_TOKEN: ${{ secrets.E2E_OPENAI_TOKEN }}
           ANTHROPIC_TOKEN: ${{ secrets.E2E_ANTHROPIC_TOKEN }}
+          # Moonshot AI platform key (platform.kimi.ai); drives both Kimi wire
+          # shapes (OpenAI /v1 and Anthropic /anthropic) through kimi_api.
+          KIMI_TOKEN: ${{ secrets.E2E_KIMI_TOKEN }}
           VERCEL_URL: ${{ secrets.E2E_VERCEL_URL }}
           VERCEL_TOKEN: ${{ secrets.E2E_VERCEL_TOKEN }}
           OPENROUTER_URL: ${{ secrets.E2E_OPENROUTER_URL }}
@@ -59,6 +69,8 @@ jobs:
           CLOUDFLARE_TOKEN: ${{ secrets.E2E_CLOUDFLARE_TOKEN }}
           AWS_BEARER_TOKEN_BEDROCK: ${{ secrets.E2E_AWS_BEARER_TOKEN_BEDROCK }}
           AWS_REGION: ${{ secrets.E2E_AWS_REGION }}
+          # Bedrock model override: dispatch input wins, then the repo variable, else the test default.
+          AWS_BEDROCK_MODEL: ${{ inputs.bedrock_model || vars.E2E_AWS_BEDROCK_MODEL }}
           # Vertex (Anthropic-on-Vertex): SA + project required; region defaults
           # to "global", model to a pinned claude snapshot.
           GOOGLE_VERTEX_SA_BASE64: ${{ secrets.E2E_GOOGLE_VERTEX_SA_BASE64 }}

.github/workflows/frontend-ui.yml

@@ -86,7 +86,7 @@ jobs:
             ${{ runner.os }}-pnpm-
 
       - name: Install dependencies
-        run: pnpm install --frozen-lockfile
+        run: pnpm install --frozen-lockfile --ignore-scripts
 
       - name: Generate Wails bindings
         run: pnpm run bindings

.github/workflows/golangci-lint.yml

@@ -45,7 +45,7 @@ jobs:
             display_name: Linux
     name: ${{ matrix.display_name }}
     runs-on: ${{ matrix.os }}
-    timeout-minutes: 15
+    timeout-minutes: 25
     steps:
       - name: Checkout code
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
@@ -79,4 +79,4 @@ jobs:
           skip-cache: true
           skip-save-cache: true
           cache-invalidation-interval: 0
-          args: --timeout=12m
+          args: --timeout=20m

client/iface/iface_test.go

@@ -464,6 +464,8 @@ func Test_RemovePeer(t *testing.T) {
 }
 
 func Test_ConnectPeers(t *testing.T) {
+	t.Setenv("NB_DISABLE_EBPF_WG_PROXY", "true")
+
 	peer1ifaceName := fmt.Sprintf("utun%d", WgIntNumber+400)
 	peer1wgIP := netip.MustParsePrefix("10.99.99.17/30")
 	peer1Key, _ := wgtypes.GeneratePrivateKey()
@@ -505,12 +507,8 @@ func Test_ConnectPeers(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	localIP, err := getLocalIP()
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	peer1endpoint, err := net.ResolveUDPAddr("udp", fmt.Sprintf("%s:%d", localIP, peer1wgPort))
+	localIP1 := "127.0.0.1"
+	peer1endpoint, err := net.ResolveUDPAddr("udp", fmt.Sprintf("%s:%d", localIP1, peer1wgPort))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -546,7 +544,8 @@ func Test_ConnectPeers(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	peer2endpoint, err := net.ResolveUDPAddr("udp", fmt.Sprintf("%s:%d", localIP, peer2wgPort))
+	localIP2 := "127.0.0.1"
+	peer2endpoint, err := net.ResolveUDPAddr("udp", fmt.Sprintf("%s:%d", localIP2, peer2wgPort))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -569,17 +568,17 @@ func Test_ConnectPeers(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	// todo: investigate why in some tests execution we need 30s
+	// The peers use userspace WireGuard (stdnet transport). A tight busy-loop
+	// here starves the wireguard-go goroutines that process the handshake, so
+	// poll on a ticker instead and yield the CPU between checks. WireGuard also
+	// only retries a lost handshake initiation every REKEY_TIMEOUT (5s), which
+	// is why the overall wait can occasionally stretch to tens of seconds.
 	timeout := 30 * time.Second
 	timeoutChannel := time.After(timeout)
+	ticker := time.NewTicker(500 * time.Millisecond)
+	defer ticker.Stop()
 
 	for {
-		select {
-		case <-timeoutChannel:
-			t.Fatalf("waiting for peer handshake timeout after %s", timeout.String())
-		default:
-		}
-
 		peer, gpErr := getPeer(peer1ifaceName, peer2Key.PublicKey().String())
 		if gpErr != nil {
 			t.Fatal(gpErr)
@@ -588,6 +587,12 @@ func Test_ConnectPeers(t *testing.T) {
 			t.Log("peers successfully handshake")
 			break
 		}
+
+		select {
+		case <-timeoutChannel:
+			t.Fatalf("waiting for peer handshake timeout after %s", timeout.String())
+		case <-ticker.C:
+		}
 	}
 
 }
@@ -615,28 +620,3 @@ func getPeer(ifaceName, peerPubKey string) (wgtypes.Peer, error) {
 	}
 	return wgtypes.Peer{}, fmt.Errorf("peer not found")
 }
-
-func getLocalIP() (string, error) {
-	// Get all interfaces
-	addrs, err := net.InterfaceAddrs()
-	if err != nil {
-		return "", err
-	}
-
-	for _, addr := range addrs {
-		ipNet, ok := addr.(*net.IPNet)
-		if !ok {
-			continue
-		}
-		if ipNet.IP.IsLoopback() {
-			continue
-		}
-
-		if ipNet.IP.To4() == nil {
-			continue
-		}
-		return ipNet.IP.String(), nil
-	}
-
-	return "", fmt.Errorf("no local IP found")
-}

client/internal/auth/auth.go

@@ -351,6 +351,7 @@ func (a *Auth) setSystemInfoFlags(info *system.Info) {
 		a.config.BlockLANAccess,
 		a.config.BlockInbound,
 		a.config.DisableIPv6,
+		a.config.SyncMessageVersion,
 		a.config.EnableSSHRoot,
 		a.config.EnableSSHSFTP,
 		a.config.EnableSSHLocalPortForwarding,

client/internal/auth/sessionwatch/watcher.go

@@ -24,11 +24,7 @@ import (
 )
 
 const (
-	// Skew tolerates a small clock difference between the management
-	// server and this peer before treating a deadline as "in the past".
-	// Slightly above typical NTP drift; tight enough that the UI doesn't
-	// paint a stale expiry as if it were valid.
-	Skew = 30 * time.Second
+	maxPastHorizon = 30 * 24 * time.Hour
 
 	// maxDeadlineHorizon caps how far in the future an accepted deadline
 	// can sit. A timestamp beyond this is almost certainly a protocol
@@ -57,7 +53,7 @@ var (
 	ErrDeadlineTooFarFuture = errors.New("session deadline too far in the future")
 
 	// ErrDeadlineInPast is returned by Update when the supplied deadline
-	// is more than Skew in the past.
+	// is more than maxPastHorizon in the past.
 	ErrDeadlineInPast = errors.New("session deadline in the past")
 )
 
@@ -66,15 +62,14 @@ var (
 // for deadline change/clear, PublishEvent for the two warnings); tests pass
 // a fake recorder so the same surface is observable without an engine.
 //
-// The watcher is the single owner of the deadline propagated to the
-// recorder: every set, clear, sanity-check rejection and Close routes the
-// value through SetSessionExpiresAt, so the SubscribeStatus snapshot the UI
-// reads can never drift from the watcher's timer state. (SetSessionExpiresAt
-// fans out its own state-change notification, so no separate notify is
-// needed.) The recorder is server-scoped and outlives this engine-scoped
-// watcher — without the Close-time clear a teardown (Down, or the Down+Up of
-// a profile switch) would leave the next session showing the previous one's
-// stale "expires in" value.
+// While the watcher runs, it owns the deadline propagated to the recorder:
+// every set, clear and sanity-check rejection routes the value through
+// SetSessionExpiresAt, so the SubscribeStatus snapshot the UI reads can
+// never drift from the watcher's timer state. (SetSessionExpiresAt fans
+// out its own state-change notification, so no separate notify is needed.)
+// The recorder is server-scoped and outlives this engine-scoped watcher;
+// Close deliberately leaves the recorder value in place so transient engine
+// restarts don't blank it — the client run loop clears it on real teardown.
 //
 // PublishEvent's signature mirrors peer.Status.PublishEvent: the watcher
 // composes the metadata internally so the wire format (MetaSession*) is
@@ -135,10 +130,13 @@ func NewWithLeads(lead, final time.Duration, recorder StatusRecorder) *Watcher {
 // was disabled).
 //
 // Same-value updates are no-ops. A different non-zero value cancels any
-// pending timer, resets the "already fired" guard, and arms a new one.
+// pending timer, resets the "already fired" guards, and — when the
+// deadline lies in the future — arms fresh warning timers. A deadline
+// already in the past (within maxPastHorizon) is recorded as-is with no
+// timers: the session has expired and consumers render it that way.
 //
 // Returns one of the sentinel Err* values when the deadline fails the
-// sanity checks (pre-epoch, far future, or in the past beyond Skew).
+// sanity checks (pre-epoch, far future, or past beyond maxPastHorizon).
 // In every error case the watcher first clears its state so it stays
 // consistent with what the caller will push into its other sinks (e.g.
 // applySessionDeadline forces a zero deadline into the status recorder
@@ -163,7 +161,7 @@ func (w *Watcher) Update(deadline time.Time) error {
 	case deadline.After(now.Add(maxDeadlineHorizon)):
 		w.clearLocked()
 		return fmt.Errorf("%w: %v", ErrDeadlineTooFarFuture, deadline)
-	case deadline.Before(now.Add(-Skew)):
+	case deadline.Before(now.Add(-maxPastHorizon)):
 		w.clearLocked()
 		return fmt.Errorf("%w: %v (now=%v)", ErrDeadlineInPast, deadline, now)
 	}
@@ -183,7 +181,9 @@ func (w *Watcher) Update(deadline time.Time) error {
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
 
-	w.armTimerLocked(deadline)
+	if deadline.After(now) {
+		w.armTimerLocked(deadline)
+	}
 	recorder := w.recorder
 	w.mu.Unlock()
 	if recorder != nil {
@@ -227,30 +227,25 @@ func (w *Watcher) Dismiss() {
 	log.Infof("auth session final-warning dismissed for deadline %s", w.current.Format(time.RFC3339))
 }
 
-// Close stops any pending timer and drops the deadline on the status
-// recorder. Update calls after Close are ignored. Clearing the recorder
-// here is what keeps a teardown (Down, or the Down+Up of a profile switch)
-// from leaving the next session showing this one's stale "expires in"
-// value — the recorder is server-scoped and outlives this engine-scoped
-// watcher, so nothing else drops the anchor on teardown.
+// Close stops any pending timer. Update calls after Close are ignored.
+// The recorder keeps its deadline: the watcher is engine-scoped and closes
+// on every engine restart (network change, sleep/wake, stream errors)
+// while the SSO deadline stays valid across those, so clearing here would
+// blank the UI's "expires in" row on every transient reconnect. The
+// client run loop clears the server-scoped recorder when it exits for
+// real (Down, profile switch, permanent login failure).
 func (w *Watcher) Close() {
 	w.mu.Lock()
+	defer w.mu.Unlock()
 	if w.closed {
-		w.mu.Unlock()
 		return
 	}
 	w.closed = true
 	w.stopTimerLocked()
-	hadDeadline := !w.current.IsZero()
 	w.current = time.Time{}
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
-	recorder := w.recorder
-	w.mu.Unlock()
-	if recorder != nil && hadDeadline {
-		recorder.SetSessionExpiresAt(time.Time{})
-	}
 }
 
 // clearLocked drops the tracked deadline and notifies the recorder so

client/internal/auth/sessionwatch/watcher_test.go

@@ -224,11 +224,13 @@ func TestNewDeadlineCancelsPriorTimer(t *testing.T) {
 
 func TestRefreshAfterFireArmsNewWarning(t *testing.T) {
 	r := &fakeRecorder{}
-	lead := 30 * time.Millisecond
+	lead := 150 * time.Millisecond
 	w := newWatcher(lead, r)
 	defer w.Close()
 
-	first := time.Now().Add(50 * time.Millisecond)
+	// Warning fires ~20ms in; the deadline itself stays 150ms away so the
+	// replacement below lands well before it.
+	first := time.Now().Add(170 * time.Millisecond)
 	_ = w.Update(first)
 
 	// Wait for stateChange + warning of the first cycle.
@@ -306,7 +308,29 @@ func TestUpdateRejectsTooFarFuture(t *testing.T) {
 	}
 }
 
-func TestUpdateInPastClearsDeadline(t *testing.T) {
+func TestUpdateRecentPastRecordedAsExpired(t *testing.T) {
+	r := &fakeRecorder{}
+	w := newWatcher(50*time.Millisecond, r)
+	defer w.Close()
+
+	d := time.Now().Add(-1 * time.Hour)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("recent-past Update should succeed, got %v", err)
+	}
+	if !w.Deadline().Equal(d) {
+		t.Fatalf("expected deadline to be recorded, got %v want %v", w.Deadline(), d)
+	}
+	if got := r.deadline(); !got.Equal(d) {
+		t.Fatalf("recorder deadline = %v, want %v", got, d)
+	}
+
+	time.Sleep(80 * time.Millisecond)
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("no warning events may fire for an already-past deadline, got %+v", r.snapshot())
+	}
+}
+
+func TestUpdateAncientPastRejected(t *testing.T) {
 	r := &fakeRecorder{}
 	w := newWatcher(50*time.Millisecond, r)
 	defer w.Close()
@@ -318,12 +342,12 @@ func TestUpdateInPastClearsDeadline(t *testing.T) {
 	// Drain the stateChange from the seed.
 	waitForEvents(t, r, 1)
 
-	err := w.Update(time.Now().Add(-1 * time.Hour))
+	err := w.Update(time.Now().Add(-31 * 24 * time.Hour))
 	if !errors.Is(err, ErrDeadlineInPast) {
 		t.Fatalf("want ErrDeadlineInPast, got %v", err)
 	}
 	if !w.Deadline().IsZero() {
-		t.Fatalf("in-past update must clear the deadline, got %v", w.Deadline())
+		t.Fatalf("rejected ancient-past update must clear the deadline, got %v", w.Deadline())
 	}
 	events := waitForEvents(t, r, 2)
 	if events[1].kind != stateChange {
@@ -331,39 +355,25 @@ func TestUpdateInPastClearsDeadline(t *testing.T) {
 	}
 }
 
-func TestUpdateWithinSkewAccepted(t *testing.T) {
-	r := &fakeRecorder{}
-	w := newWatcher(50*time.Millisecond, r)
-	defer w.Close()
-
-	// 5 seconds in the past is within the 30s Skew tolerance — accept it.
-	d := time.Now().Add(-5 * time.Second)
-	if err := w.Update(d); err != nil {
-		t.Fatalf("within-skew Update should succeed, got %v", err)
-	}
-	if !w.Deadline().Equal(d) {
-		t.Fatalf("expected deadline to be applied, got %v want %v", w.Deadline(), d)
-	}
-}
-
 func TestCloseSilencesUpdates(t *testing.T) {
 	r := &fakeRecorder{}
 	w := newWatcher(50*time.Millisecond, r)
 	w.Close()
 
-	_ = w.Update(time.Now().Add(time.Hour))
-
-	time.Sleep(20 * time.Millisecond)
+	if err := w.Update(time.Now().Add(time.Hour)); err != nil {
+		t.Fatalf("Update after Close: want nil, got %v", err)
+	}
 	if got := r.snapshot(); len(got) != 0 {
 		t.Fatalf("expected no events after Close, got %+v", got)
 	}
 }
 
-// TestCloseClearsRecorderDeadline pins the profile-switch fix: a watcher
-// holding a live deadline must zero the recorder on Close so the next
-// engine's watcher (and the UI reading the shared server-scoped recorder)
-// doesn't start out showing the previous session's stale "expires in".
-func TestCloseClearsRecorderDeadline(t *testing.T) {
+// TestCloseKeepsRecorderDeadline pins the reconnect-flap fix: the watcher
+// closes on every engine restart (network change, sleep/wake) while the
+// SSO deadline stays valid across those, so Close must leave the
+// server-scoped recorder's value in place. The client run loop clears the
+// recorder when it exits for real.
+func TestCloseKeepsRecorderDeadline(t *testing.T) {
 	r := &fakeRecorder{}
 	w := newWatcher(time.Hour, r)
 
@@ -377,8 +387,8 @@ func TestCloseClearsRecorderDeadline(t *testing.T) {
 
 	w.Close()
 
-	if got := r.deadline(); !got.IsZero() {
-		t.Fatalf("recorder deadline after Close = %v, want zero", got)
+	if got := r.deadline(); !got.Equal(d) {
+		t.Fatalf("recorder deadline after Close = %v, want %v", got, d)
 	}
 }
 

client/internal/conn_mgr.go

@@ -34,6 +34,8 @@ const (
 // - Handling connection establishment based on peer signaling
 //
 // The implementation is not thread-safe; it is protected by engine.syncMsgMux.
+// The only exception is ActivatePeer, which is safe for concurrent use so the
+// DNS warm-up path can call it without contending on the engine mutex.
 type ConnMgr struct {
 	peerStore        *peerstore.Store
 	statusRecorder   *peer.Status
@@ -42,12 +44,26 @@ type ConnMgr struct {
 	rosenpassEnabled bool
 
 	lazyConnMgr *manager.Manager
+	// lazyConnMgrMu guards the lazyConnMgr pointer for readers outside the
+	// engine loop (ActivatePeer). Writers hold it in addition to
+	// engine.syncMsgMux; all other reads stay under engine.syncMsgMux only.
+	lazyConnMgrMu sync.RWMutex
+
+	// reconcileRoutedIPs re-applies a peer's routed allowed IPs after its lazy wake endpoint is
+	// (re)armed (Mode A at arm time). Injected by the engine; nil disables the reconcile.
+	reconcileRoutedIPs func(peerKey string) error
 
 	wg            sync.WaitGroup
 	lazyCtx       context.Context
 	lazyCtxCancel context.CancelFunc
 }
 
+// SetRoutedIPsReconciler injects the callback used to re-apply a peer's routed allowed IPs when
+// its lazy wake endpoint is (re)armed. Must be called before the lazy manager starts.
+func (e *ConnMgr) SetRoutedIPsReconciler(fn func(peerKey string) error) {
+	e.reconcileRoutedIPs = fn
+}
+
 func NewConnMgr(engineConfig *EngineConfig, statusRecorder *peer.Status, peerStore *peerstore.Store, iface lazyconn.WGIface) *ConnMgr {
 	e := &ConnMgr{
 		peerStore:        peerStore,
@@ -238,12 +254,20 @@ func (e *ConnMgr) RemovePeerConn(peerKey string) {
 	conn.Log.Infof("removed peer from lazy conn manager")
 }
 
+// ActivatePeer wakes an idle lazy connection. Unlike the rest of ConnMgr it is
+// safe for concurrent use: the lazy manager pointer is read under lazyConnMgrMu
+// and the manager itself is internally synchronized, so callers outside the
+// engine loop (DNS warm-up) do not need engine.syncMsgMux.
 func (e *ConnMgr) ActivatePeer(ctx context.Context, conn *peer.Conn) {
-	if !e.isStartedWithLazyMgr() {
+	e.lazyConnMgrMu.RLock()
+	lazyConnMgr := e.lazyConnMgr
+	started := lazyConnMgr != nil && e.lazyCtxCancel != nil
+	e.lazyConnMgrMu.RUnlock()
+	if !started {
 		return
 	}
 
-	if found := e.lazyConnMgr.ActivatePeer(conn.GetKey()); found {
+	if found := lazyConnMgr.ActivatePeer(conn.GetKey()); found {
 		if err := conn.Open(ctx); err != nil {
 			conn.Log.Errorf("failed to open connection: %v", err)
 		}
@@ -268,16 +292,22 @@ func (e *ConnMgr) Close() {
 
 	e.lazyCtxCancel()
 	e.wg.Wait()
+
+	e.lazyConnMgrMu.Lock()
 	e.lazyConnMgr = nil
+	e.lazyConnMgrMu.Unlock()
 }
 
 func (e *ConnMgr) initLazyManager(engineCtx context.Context) {
 	cfg := manager.Config{
 		InactivityThreshold: inactivityThresholdEnv(),
+		ReconcileAllowedIPs: e.reconcileRoutedIPs,
 	}
-	e.lazyConnMgr = manager.NewManager(cfg, engineCtx, e.peerStore, e.iface)
 
+	e.lazyConnMgrMu.Lock()
+	e.lazyConnMgr = manager.NewManager(cfg, engineCtx, e.peerStore, e.iface)
 	e.lazyCtx, e.lazyCtxCancel = context.WithCancel(engineCtx)
+	e.lazyConnMgrMu.Unlock()
 
 	e.wg.Add(1)
 	go func() {
@@ -316,7 +346,10 @@ func (e *ConnMgr) closeManager(ctx context.Context) {
 
 	e.lazyCtxCancel()
 	e.wg.Wait()
+
+	e.lazyConnMgrMu.Lock()
 	e.lazyConnMgr = nil
+	e.lazyConnMgrMu.Unlock()
 
 	for _, peerID := range e.peerStore.PeersPubKey() {
 		e.peerStore.PeerConnOpen(ctx, peerID)

client/internal/conn_mgr_test.go

@@ -1,10 +1,21 @@
 package internal
 
 import (
+	"context"
+	"net"
+	"net/netip"
 	"os"
+	"sync"
 	"testing"
+	"time"
 
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/internal/lazyconn"
+	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/client/internal/peerstore"
+	"github.com/netbirdio/netbird/monotime"
 )
 
 func TestResolveLazyForce(t *testing.T) {
@@ -38,3 +49,58 @@ func TestResolveLazyForce(t *testing.T) {
 		})
 	}
 }
+
+type mockLazyWGIface struct{}
+
+func (mockLazyWGIface) RemovePeer(string) error { return nil }
+func (mockLazyWGIface) UpdatePeer(string, []netip.Prefix, time.Duration, *net.UDPAddr, *wgtypes.Key) error {
+	return nil
+}
+func (mockLazyWGIface) IsUserspaceBind() bool                    { return false }
+func (mockLazyWGIface) Address() wgaddr.Address                  { return wgaddr.Address{} }
+func (mockLazyWGIface) LastActivities() map[string]monotime.Time { return nil }
+func (mockLazyWGIface) MTU() uint16                              { return 1280 }
+
+// TestConnMgr_ActivatePeerConcurrentWithLifecycle exercises ActivatePeer from
+// non-engine goroutines (the DNS warm-up path) racing the manager lifecycle,
+// which stays on the engine loop. Run with -race: it fails if ActivatePeer
+// still requires engine.syncMsgMux for safety.
+func TestConnMgr_ActivatePeerConcurrentWithLifecycle(t *testing.T) {
+	t.Setenv(lazyconn.EnvLazyConn, "on")
+
+	status := peer.NewRecorder("https://mgm")
+	store := peerstore.NewConnStore()
+	connMgr := NewConnMgr(&EngineConfig{}, status, store, mockLazyWGIface{})
+
+	conn := newTestPeerConn(t, "peerA")
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	connMgr.Start(ctx)
+
+	done := make(chan struct{})
+	var wg sync.WaitGroup
+	for range 4 {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for {
+				select {
+				case <-done:
+					return
+				default:
+					connMgr.ActivatePeer(ctx, conn)
+				}
+			}
+		}()
+	}
+
+	// Let the activators spin against the started manager, then tear it down
+	// underneath them and let them spin against the stopped manager.
+	time.Sleep(100 * time.Millisecond)
+	connMgr.Close()
+	time.Sleep(50 * time.Millisecond)
+
+	close(done)
+	wg.Wait()
+}

client/internal/connect.go

@@ -257,7 +257,10 @@ func (c *ConnectClient) run(mobileDependency MobileDependency, runningChan chan
 		log.Errorf("failed to clean up temporary installer file: %v", err)
 	}
 
-	defer c.statusRecorder.ClientStop()
+	defer func() {
+		c.statusRecorder.SetSessionExpiresAt(time.Time{})
+		c.statusRecorder.ClientStop()
+	}()
 	operation := func() error {
 		// if context cancelled we not start new backoff cycle
 		if c.ctx.Err() != nil {
@@ -618,6 +621,7 @@ func createEngineConfig(key wgtypes.Key, config *profilemanager.Config, peerConf
 		BlockLANAccess:      config.BlockLANAccess,
 		BlockInbound:        config.BlockInbound,
 		DisableIPv6:         config.DisableIPv6,
+		SyncMessageVersion:  config.SyncMessageVersion,
 
 		LazyConnection: lazyconn.ParseState(config.LazyConnection),
 
@@ -693,6 +697,7 @@ func loginToManagement(ctx context.Context, client mgm.Client, pubSSHKey []byte,
 		config.BlockLANAccess,
 		config.BlockInbound,
 		config.DisableIPv6,
+		config.SyncMessageVersion,
 		config.EnableSSHRoot,
 		config.EnableSSHSFTP,
 		config.EnableSSHLocalPortForwarding,

client/internal/debug/debug.go

@@ -676,6 +676,7 @@ func (g *BundleGenerator) addCommonConfigFields(configContent *strings.Builder)
 	configContent.WriteString(fmt.Sprintf("BlockLANAccess: %v\n", g.internalConfig.BlockLANAccess))
 	configContent.WriteString(fmt.Sprintf("BlockInbound: %v\n", g.internalConfig.BlockInbound))
 	configContent.WriteString(fmt.Sprintf("DisableIPv6: %v\n", g.internalConfig.DisableIPv6))
+	configContent.WriteString(fmt.Sprintf("SyncMessageVersion: %v\n", g.internalConfig.SyncMessageVersion))
 
 	if g.internalConfig.DisableNotifications != nil {
 		configContent.WriteString(fmt.Sprintf("DisableNotifications: %v\n", *g.internalConfig.DisableNotifications))

client/internal/debug/debug_test.go

@@ -887,6 +887,8 @@ func TestAddConfig_AllFieldsCovered(t *testing.T) {
 		ClientCertKeyPath:             "/tmp/key",
 		LazyConnection:                "on",
 		MTU:                           1280,
+		DisableIPv6:                   true,
+		SyncMessageVersion:            func(v int) *int { return &v }(1),
 	}
 
 	for _, anonymize := range []bool{false, true} {

client/internal/dns/handler_chain.go

@@ -3,6 +3,7 @@ package dns
 import (
 	"context"
 	"fmt"
+	"maps"
 	"math"
 	"net"
 	"slices"
@@ -31,6 +32,26 @@ type SubdomainMatcher interface {
 	MatchSubdomains() bool
 }
 
+// responseMeta holds the annotations handlers attach to a request to explain the
+// response the chain ends up writing. It survives a deferral, so an answer that
+// did not come from the handler that owns the name still says who stepped aside
+// and why.
+type responseMeta map[resutil.MetaKey]string
+
+// format renders the annotations for the response log line. The order is stable
+// so the same event reads the same way every time; a map's own order is not.
+func (m responseMeta) format() string {
+	if len(m) == 0 {
+		return ""
+	}
+
+	var b strings.Builder
+	for _, k := range slices.Sorted(maps.Keys(m)) {
+		b.WriteString(" " + string(k) + "=" + m[k])
+	}
+	return b.String()
+}
+
 type HandlerEntry struct {
 	Handler         dns.Handler
 	Priority        int
@@ -52,8 +73,19 @@ type ResponseWriterChain struct {
 	origPattern    string
 	requestID      string
 	shouldContinue bool
-	response       *dns.Msg
-	meta           map[string]string
+	// softNegative suppresses a poisoning negative verdict for this request. A
+	// handler that owns the name but cannot answer this query type sets it
+	// before deferring, and it stays set for every handler that runs after.
+	softNegative bool
+	// clientHasEdns records whether the original query advertised EDNS0, taken
+	// before any handler ran: handlers add EDNS0 to the query they forward
+	// upstream, so the message itself no longer answers the question later.
+	clientHasEdns bool
+	response      *dns.Msg
+	// meta is handed to the next handler when this one defers, so the same map
+	// outlives the writer that created it. Handlers must only set metadata from
+	// within ServeDNS, never from a goroutine that outlives the call.
+	meta responseMeta
 }
 
 // RequestID returns the request ID for tracing
@@ -62,26 +94,64 @@ func (w *ResponseWriterChain) RequestID() string {
 }
 
 // SetMeta sets a metadata key-value pair for logging
-func (w *ResponseWriterChain) SetMeta(key, value string) {
+func (w *ResponseWriterChain) SetMeta(key resutil.MetaKey, value string) {
 	if w.meta == nil {
-		w.meta = make(map[string]string)
+		w.meta = make(responseMeta)
 	}
 	w.meta[key] = value
 }
 
+// RequestSoftNegative marks the request so a downstream NXDOMAIN is turned into
+// NODATA before it reaches the client. Set by a handler that defers a query for
+// a name it owns but a type it cannot resolve.
+func (w *ResponseWriterChain) RequestSoftNegative() {
+	w.softNegative = true
+}
+
 func (w *ResponseWriterChain) WriteMsg(m *dns.Msg) error {
 	// Check if this is a continue signal (NXDOMAIN with Zero bit set)
 	if m.Rcode == dns.RcodeNameError && m.MsgHdr.Zero {
 		w.shouldContinue = true
 		return nil
 	}
+	if w.softNegative && m.Rcode == dns.RcodeNameError {
+		m = softenNegative(m, w.clientHasEdns)
+		w.SetMeta(resutil.MetaKeySoftened, "nxdomain->nodata")
+	}
 	w.response = m
 	if m.MsgHdr.Truncated {
-		w.SetMeta("truncated", "true")
+		w.SetMeta(resutil.MetaKeyTruncated, "true")
 	}
 	return w.ResponseWriter.WriteMsg(m)
 }
 
+// softenNegative downgrades an NXDOMAIN to NODATA for a request a handler that
+// owns the name deferred. NXDOMAIN is cached for the name and every type below
+// it, so a resolver that has never heard of a routed name would take out the
+// record types the route does serve; NODATA is cached for this name and type
+// only. The authority section goes with it: the negative TTL of a zone we just
+// overruled does not apply, and RFC 2308 keeps a negative answer that carries no
+// SOA out of caches altogether, so the rewrite cannot outlive the route.
+//
+// The rewrite is ours, not the answering resolver's, so an EDNS0 client is told
+// as much: the reply we hand back travels a path no capture on this host sees,
+// and an empty answer is otherwise indistinguishable from a real one. Returns a
+// copy so the answering handler keeps whatever it may hold on to.
+func softenNegative(m *dns.Msg, clientHasEdns bool) *dns.Msg {
+	out := m.Copy()
+	out.Rcode = dns.RcodeSuccess
+	out.Ns = nil
+
+	if clientHasEdns {
+		resutil.AttachEDE(out, resutil.EDENetbirdSoftenedNegative,
+			"netbird: name is served locally, NXDOMAIN from the fallthrough resolver suppressed")
+	} else {
+		resutil.StripOPT(out)
+	}
+
+	return out
+}
+
 func NewHandlerChain() *HandlerChain {
 	return &HandlerChain{
 		handlers: make([]HandlerEntry, 0),
@@ -223,6 +293,19 @@ func (c *HandlerChain) dispatch(w dns.ResponseWriter, r *dns.Msg, maxPriority in
 	handlers := slices.Clone(c.handlers)
 	c.mu.RUnlock()
 
+	// Carried across deferrals: once a handler that owns the name defers, no
+	// handler after it may answer with a verdict that poisons the name. The
+	// metadata of a handler that stepped aside is carried too, so the response
+	// log line explains an answer that did not come from the handler that owns
+	// the name.
+	var softNegative bool
+	var carried responseMeta
+
+	// Taken before any handler runs: handlers advertise EDNS0 on the query they
+	// forward upstream, so afterwards the message no longer tells us whether the
+	// client did.
+	clientHasEdns := r.IsEdns0() != nil
+
 	// Try handlers in priority order
 	for _, entry := range handlers {
 		if entry.Priority > maxPriority {
@@ -245,11 +328,16 @@ func (c *HandlerChain) dispatch(w dns.ResponseWriter, r *dns.Msg, maxPriority in
 			ResponseWriter: w,
 			origPattern:    entry.OrigPattern,
 			requestID:      requestID,
+			softNegative:   softNegative,
+			clientHasEdns:  clientHasEdns,
+			meta:           carried,
 		}
 		entry.Handler.ServeDNS(chainWriter, r)
 
 		// If handler wants to continue, try next handler
 		if chainWriter.shouldContinue {
+			softNegative = softNegative || chainWriter.softNegative
+			carried = chainWriter.meta
 			if entry.Priority != PriorityMgmtCache {
 				logger.Tracef("handler requested continue for domain=%s", qname)
 			}
@@ -265,30 +353,59 @@ func (c *HandlerChain) dispatch(w dns.ResponseWriter, r *dns.Msg, maxPriority in
 		qname, dns.TypeToString[question.Qtype], dns.ClassToString[question.Qclass])
 	resp := &dns.Msg{}
 	resp.SetRcode(r, dns.RcodeRefused)
+	// A handler that owns the name deferred and nothing below it could answer
+	// (a client with no primary nameserver group). The name exists as far as
+	// this client is concerned, since the route serves its addresses, so REFUSED
+	// would contradict the route: a stub that takes it as "not served here" and
+	// retries another resolver can come back with an NXDOMAIN that takes the
+	// whole name down. Answer "no records of this type" instead, without an SOA,
+	// so RFC 2308 keeps it out of negative caches, and tell an EDNS0 client the
+	// empty answer is ours rather than a resolver's.
+	if softNegative {
+		resp.Rcode = dns.RcodeSuccess
+		if clientHasEdns {
+			resutil.AttachEDE(resp, resutil.EDENetbirdSoftenedNegative,
+				"netbird: name is served locally, no fallthrough resolver for this query type")
+		}
+		// logResponse never runs on this path, so the carried metadata is
+		// appended here or the reason for the deferral is lost in exactly the
+		// case that is hardest to diagnose.
+		logger.Tracef("no handler below the deferring one for domain=%s type=%s, answering NODATA%s",
+			qname, dns.TypeToString[question.Qtype], carried.format())
+	}
 	if err := w.WriteMsg(resp); err != nil {
 		logger.Errorf("failed to write DNS response: %v", err)
 	}
 }
 
 func (c *HandlerChain) logResponse(logger *log.Entry, cw *ResponseWriterChain, qname string, startTime time.Time) {
+	// Runs for every query, and the arguments below are not free: Len() packs
+	// the message to measure it, and formatting the answers and the metadata
+	// allocates. None of it is worth doing when the line is discarded.
+	if !log.IsLevelEnabled(log.TraceLevel) {
+		return
+	}
+
 	if cw.response == nil {
 		return
 	}
 
-	var meta string
-	for k, v := range cw.meta {
-		meta += " " + k + "=" + v
-	}
-
 	logger.Tracef("response: domain=%s rcode=%s answers=%s size=%dB%s took=%s",
 		qname, dns.RcodeToString[cw.response.Rcode], resutil.FormatAnswers(cw.response.Answer),
-		cw.response.Len(), meta, time.Since(startTime))
+		cw.response.Len(), cw.meta.format(), time.Since(startTime))
 }
 
 // ResolveInternal runs an in-process DNS query against the chain, skipping any
 // handler with priority > maxPriority. Used by internal callers (e.g. the mgmt
-// cache refresher) that must bypass themselves to avoid loops. Honors ctx
-// cancellation; on ctx.Done the dispatch goroutine is left to drain on its own
+// cache refresher) that must bypass themselves to avoid loops.
+//
+// "Nothing answered" is read off RcodeRefused, which a request soft-negatived by
+// a deferring handler never carries: it ends in an empty NOERROR instead, and
+// would look resolved. No caller can reach that today, since every handler that
+// defers sits above the maxPriority any caller passes. Lowering one below it
+// means this check needs the soft-negative case too.
+//
+// Honors ctx cancellation; on ctx.Done the dispatch goroutine is left to drain on its own
 // (bounded by the invoked handler's internal timeout).
 func (c *HandlerChain) ResolveInternal(ctx context.Context, r *dns.Msg, maxPriority int) (*dns.Msg, error) {
 	if len(r.Question) == 0 {

client/internal/dns/handler_chain_test.go

@@ -3,15 +3,19 @@ package dns_test
 import (
 	"context"
 	"net"
+	"strings"
 	"testing"
 	"time"
 
 	"github.com/miekg/dns"
+	log "github.com/sirupsen/logrus"
+	logtest "github.com/sirupsen/logrus/hooks/test"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/mock"
 	"github.com/stretchr/testify/require"
 
 	nbdns "github.com/netbirdio/netbird/client/internal/dns"
+	"github.com/netbirdio/netbird/client/internal/dns/resutil"
 	"github.com/netbirdio/netbird/client/internal/dns/test"
 )
 
@@ -1238,6 +1242,276 @@ func TestHandlerChain_ResolveInternal_HonorsContextTimeout(t *testing.T) {
 	assert.Less(t, elapsed, 500*time.Millisecond, "ResolveInternal must return shortly after ctx deadline")
 }
 
+// requestSoftNegative asks the chain to soften a negative verdict produced by
+// the handlers that run after the caller defers, reporting whether the chain
+// supports the signal. Written as a type assertion so these tests compile
+// against a chain that does not support it yet.
+func requestSoftNegative(w dns.ResponseWriter) bool {
+	sn, ok := w.(interface{ RequestSoftNegative() })
+	if ok {
+		sn.RequestSoftNegative()
+	}
+	return ok
+}
+
+// deferringHandler defers to the next handler in the chain, optionally asking
+// for the negative verdict of whatever answers instead to be softened. This is
+// what a DNS route handler does for a record type its routing peer cannot
+// resolve.
+type deferringHandler struct {
+	softNegative bool
+	called       bool
+	supported    bool
+}
+
+func (h *deferringHandler) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
+	h.called = true
+	if h.softNegative {
+		h.supported = requestSoftNegative(w)
+		// The real handler records why it stepped aside, for the response log
+		// line of whichever handler answers instead.
+		resutil.SetMeta(w, resutil.MetaKeyDeferredBy, "test handler")
+	}
+	resp := new(dns.Msg)
+	resp.SetRcode(r, dns.RcodeNameError)
+	resp.MsgHdr.Zero = true
+	_ = w.WriteMsg(resp)
+}
+
+// nxdomainHandler answers an authoritative NXDOMAIN with an SOA in the
+// authority section, the way a public resolver answers for a name that only
+// exists inside the routed network.
+type nxdomainHandler struct{}
+
+func (h *nxdomainHandler) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
+	resp := new(dns.Msg)
+	resp.SetRcode(r, dns.RcodeNameError)
+	resp.Ns = []dns.RR{&dns.SOA{
+		Hdr:     dns.RR_Header{Name: "example.com.", Rrtype: dns.TypeSOA, Class: dns.ClassINET, Ttl: 3600},
+		Ns:      "ns1.example.com.",
+		Mbox:    "hostmaster.example.com.",
+		Minttl:  3600,
+		Expire:  604800,
+		Refresh: 7200,
+		Retry:   3600,
+	}}
+	_ = w.WriteMsg(resp)
+}
+
+// TestHandlerChain_SoftNegative_DowngradesDownstreamNXDOMAIN is the whole point
+// of the soft-negative signal: a route handler may only defer a query to the
+// public chain if the answer cannot poison the routed name. NXDOMAIN is cached
+// for the name and every type under it (RFC 2308, RFC 8020), so it has to be
+// rewritten to NODATA, which is cached per name and type only.
+func TestHandlerChain_SoftNegative_DowngradesDownstreamNXDOMAIN(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+
+	route := &deferringHandler{softNegative: true}
+	chain.AddHandler("*.example.com.", route, nbdns.PriorityDNSRoute)
+	chain.AddHandler(".", &nxdomainHandler{}, nbdns.PriorityDefault)
+
+	r := new(dns.Msg)
+	r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+
+	mw := &test.MockResponseWriter{}
+	chain.ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+	resp := mw.GetLastResponse()
+	require.NotNil(t, resp, "a response must reach the client")
+	assert.True(t, route.called, "the route handler must run first")
+	require.True(t, route.supported, "the chain writer must accept a soft-negative request")
+	assert.Equal(t, dns.RcodeSuccess, resp.Rcode, "NXDOMAIN must be softened to NODATA")
+	assert.Empty(t, resp.Answer, "a softened negative carries no answer")
+	assert.Empty(t, resp.Ns, "the downstream zone's SOA must not set the negative TTL for a name we overrode")
+}
+
+// responseLineFor returns the chain's response log line for one query. Raising
+// the level to trace also unmutes whatever else is logging in this package, so
+// the line has to be picked by the name it was asked about rather than by being
+// the last one seen.
+func responseLineFor(hook *logtest.Hook, qname string) string {
+	for _, e := range hook.AllEntries() {
+		if strings.HasPrefix(e.Message, "response:") && strings.Contains(e.Message, qname) {
+			return e.Message
+		}
+	}
+	return ""
+}
+
+// TestHandlerChain_SoftNegative_IsVisibleToTheClient covers observability of the
+// rewrite. The reply we hand the application travels over loopback, which the
+// bundle capture does not see, so a softened verdict has to say so on the wire:
+// without it an empty answer is indistinguishable from a real "no such record"
+// in a dig output or a capture taken next to the application.
+func TestHandlerChain_SoftNegative_IsVisibleToTheClient(t *testing.T) {
+	// One hook for the whole test: logtest installs it on the standard logger
+	// and logrus has no way to take it off again, so a hook per subtest would
+	// leave several behind buffering every later line in the package.
+	hook := logtest.NewGlobal()
+	t.Cleanup(hook.Reset)
+
+	newChain := func() *nbdns.HandlerChain {
+		chain := nbdns.NewHandlerChain()
+		chain.AddHandler("*.example.com.", &deferringHandler{softNegative: true}, nbdns.PriorityDNSRoute)
+		chain.AddHandler(".", &nxdomainHandler{}, nbdns.PriorityDefault)
+		return chain
+	}
+
+	t.Run("EDNS0 client gets an extended error", func(t *testing.T) {
+		r := new(dns.Msg)
+		r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+		r.SetEdns0(dns.DefaultMsgSize, false)
+
+		mw := &test.MockResponseWriter{}
+		newChain().ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+		resp := mw.GetLastResponse()
+		require.NotNil(t, resp)
+		require.Equal(t, dns.RcodeSuccess, resp.Rcode)
+
+		ede, ok := resutil.ExtractEDE(resp)
+		require.True(t, ok, "a softened verdict must carry an extended DNS error")
+		assert.Equal(t, resutil.EDENetbirdSoftenedNegative, ede.InfoCode)
+		assert.Contains(t, ede.ExtraText, "netbird", "the text must name us as the source of the rewrite")
+	})
+
+	t.Run("plain client gets no OPT", func(t *testing.T) {
+		r := new(dns.Msg)
+		r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+
+		mw := &test.MockResponseWriter{}
+		newChain().ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+		resp := mw.GetLastResponse()
+		require.NotNil(t, resp)
+		assert.Nil(t, resp.IsEdns0(), "RFC 6891 forbids an OPT toward a client that did not advertise EDNS0")
+	})
+
+	// The response line is what support reads out of a debug bundle, and it now
+	// carries several annotations at once. Built from a map, their order would
+	// differ on every query, so the same event never looks the same twice.
+	t.Run("log fields keep a stable order", func(t *testing.T) {
+		const qname = "_mongodb._tcp.stable.example.com."
+
+		prev := log.GetLevel()
+		log.SetLevel(log.TraceLevel)
+		t.Cleanup(func() { log.SetLevel(prev) })
+
+		lineFor := func() string {
+			hook.Reset()
+
+			r := new(dns.Msg)
+			r.SetQuestion(qname, dns.TypeSRV)
+
+			mw := &test.MockResponseWriter{}
+			newChain().ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+			line := responseLineFor(hook, qname)
+			// The duration differs per query and is not what we compare.
+			line, _, _ = strings.Cut(line, " took=")
+			return line
+		}
+
+		first := lineFor()
+		require.NotEmpty(t, first, "the chain must log the response it wrote")
+		require.Contains(t, first, "deferred_by=", "the line must carry more than one annotation to be worth ordering")
+		require.Contains(t, first, "softened=")
+
+		for range 20 {
+			assert.Equal(t, first, lineFor(), "the same event must produce the same line")
+		}
+	})
+
+	t.Run("logged with the reason it was deferred", func(t *testing.T) {
+		const qname = "_mongodb._tcp.reason.example.com."
+
+		hook.Reset()
+
+		prev := log.GetLevel()
+		log.SetLevel(log.TraceLevel)
+		t.Cleanup(func() { log.SetLevel(prev) })
+
+		r := new(dns.Msg)
+		r.SetQuestion(qname, dns.TypeSRV)
+
+		mw := &test.MockResponseWriter{}
+		newChain().ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+		response := responseLineFor(hook, qname)
+		require.NotEmpty(t, response, "the chain must log the response it wrote")
+		assert.Contains(t, response, "softened=", "the log must show the verdict was rewritten")
+		assert.Contains(t, response, "deferred_by=", "the log must name the handler that deferred")
+	})
+}
+
+// TestHandlerChain_SoftNegative_NoHandlerBelow covers a client with no primary
+// nameserver group: the deferred query reaches the end of the chain unanswered.
+// REFUSED would say the name is not served here while the route serves its
+// addresses, and a stub that acts on that by asking elsewhere can bring back an
+// NXDOMAIN for the whole name. The answer must be an empty, uncacheable NODATA.
+func TestHandlerChain_SoftNegative_NoHandlerBelow(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+
+	route := &deferringHandler{softNegative: true}
+	chain.AddHandler("*.example.com.", route, nbdns.PriorityDNSRoute)
+
+	r := new(dns.Msg)
+	r.SetQuestion("db.example.com.", dns.TypeHTTPS)
+
+	mw := &test.MockResponseWriter{}
+	chain.ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+	resp := mw.GetLastResponse()
+	require.NotNil(t, resp, "a response must reach the client")
+	assert.Equal(t, dns.RcodeSuccess, resp.Rcode, "an unanswered soft-negative query must be NODATA, not REFUSED")
+	assert.Empty(t, resp.Answer)
+	assert.Empty(t, resp.Ns,
+		"no SOA, so RFC 2308 keeps the empty answer out of negative caches and it cannot outlive the route")
+}
+
+// TestHandlerChain_SoftNegative_KeepsRealAnswers guards the other direction:
+// softening applies to negative verdicts only. A real answer from a downstream
+// handler must reach the client untouched.
+func TestHandlerChain_SoftNegative_KeepsRealAnswers(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+
+	route := &deferringHandler{softNegative: true}
+	chain.AddHandler("*.example.com.", route, nbdns.PriorityDNSRoute)
+	chain.AddHandler(".", &answeringHandler{name: "public", ip: "203.0.113.10"}, nbdns.PriorityDefault)
+
+	r := new(dns.Msg)
+	r.SetQuestion("db.example.com.", dns.TypeA)
+
+	mw := &test.MockResponseWriter{}
+	chain.ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+	resp := mw.GetLastResponse()
+	require.NotNil(t, resp, "a response must reach the client")
+	assert.Equal(t, dns.RcodeSuccess, resp.Rcode)
+	require.Len(t, resp.Answer, 1, "the downstream answer must pass through")
+}
+
+// TestHandlerChain_NXDOMAINPreservedWithoutSoftNegative makes sure the
+// softening is opt-in: an ordinary chain continuation still yields NXDOMAIN, so
+// genuine non-existence keeps being reported.
+func TestHandlerChain_NXDOMAINPreservedWithoutSoftNegative(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+
+	route := &deferringHandler{}
+	chain.AddHandler("*.example.com.", route, nbdns.PriorityDNSRoute)
+	chain.AddHandler(".", &nxdomainHandler{}, nbdns.PriorityDefault)
+
+	r := new(dns.Msg)
+	r.SetQuestion("nope.example.com.", dns.TypeA)
+
+	mw := &test.MockResponseWriter{}
+	chain.ServeDNS(&nbdns.ResponseWriterChain{ResponseWriter: mw}, r)
+
+	resp := mw.GetLastResponse()
+	require.NotNil(t, resp, "a response must reach the client")
+	assert.Equal(t, dns.RcodeNameError, resp.Rcode, "without the signal a real NXDOMAIN must survive")
+}
+
 func TestHandlerChain_HasRootHandlerAtOrBelow(t *testing.T) {
 	chain := nbdns.NewHandlerChain()
 	h := &answeringHandler{name: "h", ip: "10.0.0.1"}

client/internal/dns/local/local.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net"
 	"net/netip"
+	"os"
 	"slices"
 	"strings"
 	"sync"
@@ -36,7 +37,43 @@ type resolver interface {
 // record is left alone (it points at something outside our mesh, e.g.
 // a non-peer upstream).
 type PeerConnectivity interface {
-	IsConnectedByIP(ip string) (known, connected bool)
+	IsConnectedByIP(ip netip.Addr) (known, connected bool)
+}
+
+// PeerActivator wakes lazy-connection peers on demand. The local resolver calls
+// it with the tunnel IPs an answer points at, so a peer that is idle (lazily
+// disconnected) starts connecting at DNS-resolution time rather than racing the
+// client's first request packet. nil disables warm-up.
+type PeerActivator interface {
+	// ActivatePeersByIP triggers wake-up for the peer(s) owning addrs and blocks
+	// until one is connected or ctx (a short per-query budget) expires. It is a
+	// fast no-op for unknown or already-connected addresses.
+	ActivatePeersByIP(ctx context.Context, addrs []netip.Addr)
+}
+
+const (
+	defaultLazyWarmupTimeout = 2 * time.Second
+	envLazyWarmupTimeout     = "NB_DNS_LAZY_WARMUP_TIMEOUT"
+)
+
+// lazyWarmupTimeoutFromEnv returns the per-query budget for waking a
+// lazy-connection peer a DNS answer points at. Tunable via
+// NB_DNS_LAZY_WARMUP_TIMEOUT (a Go duration). Parsed once at construction time.
+func lazyWarmupTimeoutFromEnv() time.Duration {
+	v := os.Getenv(envLazyWarmupTimeout)
+	if v == "" {
+		return defaultLazyWarmupTimeout
+	}
+	d, err := time.ParseDuration(v)
+	if err != nil {
+		log.Warnf("invalid %s value %q, using default %s: %v", envLazyWarmupTimeout, v, defaultLazyWarmupTimeout, err)
+		return defaultLazyWarmupTimeout
+	}
+	if d <= 0 {
+		log.Warnf("non-positive %s value %q, using default %s", envLazyWarmupTimeout, v, defaultLazyWarmupTimeout)
+		return defaultLazyWarmupTimeout
+	}
+	return d
 }
 
 type Resolver struct {
@@ -51,6 +88,12 @@ type Resolver struct {
 	// filter and preserves the legacy "return whatever is registered"
 	// behaviour for callers that never wire a status source.
 	peerConn PeerConnectivity
+	// peerActivator, when non-nil, is called at resolution time to warm the
+	// lazy connection to the peer(s) an answer points at. nil disables warm-up.
+	peerActivator PeerActivator
+	// warmupTimeout is the per-query budget for the lazy-connection warm-up
+	// wait, resolved from the environment once at construction time.
+	warmupTimeout time.Duration
 
 	ctx    context.Context
 	cancel context.CancelFunc
@@ -59,11 +102,12 @@ type Resolver struct {
 func NewResolver() *Resolver {
 	ctx, cancel := context.WithCancel(context.Background())
 	return &Resolver{
-		records: make(map[dns.Question][]dns.RR),
-		domains: make(map[domain.Domain]struct{}),
-		zones:   make(map[domain.Domain]bool),
-		ctx:     ctx,
-		cancel:  cancel,
+		records:       make(map[dns.Question][]dns.RR),
+		domains:       make(map[domain.Domain]struct{}),
+		zones:         make(map[domain.Domain]bool),
+		warmupTimeout: lazyWarmupTimeoutFromEnv(),
+		ctx:           ctx,
+		cancel:        cancel,
 	}
 }
 
@@ -76,6 +120,14 @@ func (d *Resolver) SetPeerConnectivity(p PeerConnectivity) {
 	d.peerConn = p
 }
 
+// SetPeerActivator wires the DNS-time lazy-connection warm-up. Pass nil to
+// disable. Safe to call multiple times; the latest value wins.
+func (d *Resolver) SetPeerActivator(a PeerActivator) {
+	d.mu.Lock()
+	defer d.mu.Unlock()
+	d.peerActivator = a
+}
+
 func (d *Resolver) MatchSubdomains() bool {
 	return true
 }
@@ -122,6 +174,9 @@ func (d *Resolver) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 	replyMessage.RecursionAvailable = true
 
 	result := d.lookupRecords(logger, question)
+	// Warm before filtering: activation flips a lazily-idle target to connected,
+	// which then lets it survive the disconnected-peer filter below.
+	d.warmLazyPeers(question, result.records)
 	result.records = d.filterDisconnectedPeerAnswers(logger, question, result.records)
 	replyMessage.Authoritative = !result.hasExternalData
 	replyMessage.Answer = result.records
@@ -495,8 +550,8 @@ func (d *Resolver) filterDisconnectedPeerAnswers(logger *log.Entry, question dns
 	kept := make([]dns.RR, 0, len(records))
 	var dropped int
 	for _, rr := range records {
-		ip := extractRecordIP(rr)
-		if ip == "" {
+		ip, ok := extractRecordAddr(rr)
+		if !ok {
 			kept = append(kept, rr)
 			continue
 		}
@@ -518,22 +573,57 @@ func (d *Resolver) filterDisconnectedPeerAnswers(logger *log.Entry, question dns
 	return kept
 }
 
-// extractRecordIP returns the dotted-decimal / colon-hex IP carried by
-// an A or AAAA record, or "" for any other record type.
-func extractRecordIP(rr dns.RR) string {
+// warmLazyPeers triggers lazy-connection wake-up for the peers a resolved
+// answer points at and waits briefly for one to connect, so the caller's first
+// request doesn't race the connection establishment. Warm-up is scoped to
+// match-only (non-authoritative) zones — the synthesized private-service zones
+// and user-created zones whose records point at specific peers. The account's
+// peer zone is authoritative, so plain peer-name lookups never trigger warm-up;
+// otherwise resolving any peer's name would wake its idle connection, defeating
+// laziness mesh-wide. No-op when no activator is wired (lazy connections
+// disabled) or the answer carries no peer IPs.
+func (d *Resolver) warmLazyPeers(question dns.Question, records []dns.RR) {
+	if len(records) < 2 {
+		return
+	}
+	d.mu.RLock()
+	activator := d.peerActivator
+	var nonAuth, found bool
+	if activator != nil {
+		nonAuth, found = d.findZone(question.Name)
+	}
+	d.mu.RUnlock()
+	if activator == nil || !found || !nonAuth {
+		return
+	}
+
+	var addrs []netip.Addr
+	for _, rr := range records {
+		if addr, ok := extractRecordAddr(rr); ok {
+			addrs = append(addrs, addr)
+		}
+	}
+	if len(addrs) == 0 {
+		return
+	}
+
+	ctx, cancel := context.WithTimeout(d.ctx, d.warmupTimeout)
+	defer cancel()
+	activator.ActivatePeersByIP(ctx, addrs)
+}
+
+// extractRecordAddr returns the IP address carried by an A or AAAA record.
+// ok is false for any other record type or a record with no address.
+func extractRecordAddr(rr dns.RR) (netip.Addr, bool) {
 	switch r := rr.(type) {
 	case *dns.A:
-		if r.A == nil {
-			return ""
-		}
-		return r.A.String()
+		addr, ok := netip.AddrFromSlice(r.A)
+		return addr.Unmap(), ok
 	case *dns.AAAA:
-		if r.AAAA == nil {
-			return ""
-		}
-		return r.AAAA.String()
+		addr, ok := netip.AddrFromSlice(r.AAAA)
+		return addr.Unmap(), ok
 	}
-	return ""
+	return netip.Addr{}, false
 }
 
 // Update replaces all zones and their records

client/internal/dns/local/local_test.go

@@ -37,8 +37,8 @@ type mockPeerConnectivity struct {
 	byIP map[string]struct{ known, connected bool }
 }
 
-func (m mockPeerConnectivity) IsConnectedByIP(ip string) (known, connected bool) {
-	v, ok := m.byIP[ip]
+func (m mockPeerConnectivity) IsConnectedByIP(ip netip.Addr) (known, connected bool) {
+	v, ok := m.byIP[ip.String()]
 	if !ok {
 		return false, false
 	}

client/internal/dns/local/warmup_test.go

@@ -0,0 +1,204 @@
+package local
+
+import (
+	"context"
+	"net"
+	"net/netip"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/miekg/dns"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/dns/test"
+	nbdns "github.com/netbirdio/netbird/dns"
+)
+
+// recordingActivator records the addresses it was asked to warm and returns
+// immediately, so ServeDNS is not blocked by the test.
+type recordingActivator struct {
+	mu     sync.Mutex
+	called bool
+	addrs  []netip.Addr
+}
+
+func (r *recordingActivator) ActivatePeersByIP(_ context.Context, addrs []netip.Addr) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.called = true
+	r.addrs = append(r.addrs, addrs...)
+}
+
+func serveA(t *testing.T, resolver *Resolver, name string) *dns.Msg {
+	t.Helper()
+	var resp *dns.Msg
+	w := &test.MockResponseWriter{WriteMsgFunc: func(m *dns.Msg) error { resp = m; return nil }}
+	resolver.ServeDNS(w, new(dns.Msg).SetQuestion(name, dns.TypeA))
+	return resp
+}
+
+// serviceZone registers rec in a match-only (non-authoritative) zone, the shape
+// the synthesized private-service zones arrive in.
+func serviceZone(t *testing.T, resolver *Resolver, zone string, records ...nbdns.SimpleRecord) {
+	t.Helper()
+	resolver.Update([]nbdns.CustomZone{{
+		Domain:           zone,
+		Records:          records,
+		NonAuthoritative: true,
+	}})
+}
+
+func TestLocalResolver_WarmsLazyPeerOnResolve(t *testing.T) {
+	// Warm-up fires only for multi-record answers (the HA / round-robin shape of
+	// the synthesized private-service zones), so register two peer targets.
+	const name = "svc.proxy.netbird.cloud."
+	recs := []nbdns.SimpleRecord{
+		{Name: name, Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.7"},
+		{Name: name, Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.8"},
+	}
+	resolver := NewResolver()
+	serviceZone(t, resolver, "proxy.netbird.cloud", recs...)
+
+	act := &recordingActivator{}
+	resolver.SetPeerActivator(act)
+
+	resp := serveA(t, resolver, name)
+	require.NotNil(t, resp, "resolver must answer")
+	require.NotEmpty(t, resp.Answer, "answer must carry the A records")
+
+	act.mu.Lock()
+	defer act.mu.Unlock()
+	assert.True(t, act.called, "activator must be invoked for a multi-record service-zone answer")
+	assert.Contains(t, act.addrs, netip.MustParseAddr("100.64.0.7"), "activator must receive the first peer IP")
+	assert.Contains(t, act.addrs, netip.MustParseAddr("100.64.0.8"), "activator must receive the second peer IP")
+}
+
+func TestLocalResolver_NoWarmupForSingleRecord(t *testing.T) {
+	// A single-record answer does not trigger warm-up; the resolver only warms
+	// multi-record answers.
+	rec := nbdns.SimpleRecord{Name: "svc.proxy.netbird.cloud.", Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.7"}
+	resolver := NewResolver()
+	serviceZone(t, resolver, "proxy.netbird.cloud", rec)
+
+	act := &recordingActivator{}
+	resolver.SetPeerActivator(act)
+
+	resp := serveA(t, resolver, rec.Name)
+	require.NotNil(t, resp, "resolver must answer")
+	require.NotEmpty(t, resp.Answer, "answer must carry the A record")
+
+	act.mu.Lock()
+	defer act.mu.Unlock()
+	assert.False(t, act.called, "activator must not be invoked for a single-record answer")
+}
+
+func TestLocalResolver_NoActivatorNoWarmup(t *testing.T) {
+	// With no activator wired the resolver behaves exactly as before.
+	rec := nbdns.SimpleRecord{Name: "svc.proxy.netbird.cloud.", Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.7"}
+	resolver := NewResolver()
+	serviceZone(t, resolver, "proxy.netbird.cloud", rec)
+
+	resp := serveA(t, resolver, rec.Name)
+	require.NotNil(t, resp, "resolver must still answer without an activator")
+	require.NotEmpty(t, resp.Answer, "answer must carry the A record")
+}
+
+func TestLocalResolver_NoWarmupForMissingRecord(t *testing.T) {
+	// A query that resolves to nothing must not invoke the activator (no IPs).
+	resolver := NewResolver()
+	serviceZone(t, resolver, "proxy.netbird.cloud",
+		nbdns.SimpleRecord{Name: "svc.proxy.netbird.cloud.", Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.7"})
+
+	act := &recordingActivator{}
+	resolver.SetPeerActivator(act)
+
+	serveA(t, resolver, "absent.proxy.netbird.cloud.")
+
+	act.mu.Lock()
+	defer act.mu.Unlock()
+	assert.False(t, act.called, "activator must not be invoked when there is no answer")
+}
+
+func TestLocalResolver_NoWarmupInAuthoritativeZone(t *testing.T) {
+	// The account's peer zone is authoritative; resolving a peer's name there
+	// must not wake its lazy connection — warm-up is scoped to match-only
+	// (non-authoritative) zones such as the synthesized private-service zones.
+	// Use a multi-record answer so the authoritative-zone scoping is the only
+	// reason warm-up is skipped, not the single-record guard.
+	const name = "peer.netbird.cloud."
+	recs := []nbdns.SimpleRecord{
+		{Name: name, Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.9"},
+		{Name: name, Type: 1, Class: nbdns.DefaultClass, TTL: 300, RData: "100.64.0.10"},
+	}
+	resolver := NewResolver()
+	resolver.Update([]nbdns.CustomZone{{
+		Domain:  "netbird.cloud",
+		Records: recs,
+	}})
+
+	act := &recordingActivator{}
+	resolver.SetPeerActivator(act)
+
+	resp := serveA(t, resolver, name)
+	require.NotNil(t, resp, "resolver must answer")
+	require.NotEmpty(t, resp.Answer, "answer must carry the A records")
+
+	act.mu.Lock()
+	defer act.mu.Unlock()
+	assert.False(t, act.called, "activator must not be invoked for authoritative-zone answers")
+}
+
+func TestLazyWarmupTimeoutFromEnv(t *testing.T) {
+	tests := []struct {
+		name   string
+		value  string
+		envSet bool
+		want   time.Duration
+	}{
+		{name: "unset uses default", want: defaultLazyWarmupTimeout},
+		{name: "valid overrides", value: "5s", envSet: true, want: 5 * time.Second},
+		{name: "invalid falls back", value: "not-a-duration", envSet: true, want: defaultLazyWarmupTimeout},
+		{name: "negative falls back", value: "-1s", envSet: true, want: defaultLazyWarmupTimeout},
+		{name: "zero falls back", value: "0s", envSet: true, want: defaultLazyWarmupTimeout},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if tt.envSet {
+				t.Setenv(envLazyWarmupTimeout, tt.value)
+			}
+			assert.Equal(t, tt.want, lazyWarmupTimeoutFromEnv())
+			assert.Equal(t, tt.want, NewResolver().warmupTimeout, "constructor must resolve the timeout once")
+		})
+	}
+}
+
+func TestExtractRecordAddr(t *testing.T) {
+	t.Run("A record yields unmapped v4", func(t *testing.T) {
+		// net.ParseIP returns the 16-byte v4-in-v6 form, the same shape
+		// miekg/dns stores after parsing an A record; the extracted address
+		// must compare equal to a plain v4 netip.Addr.
+		addr, ok := extractRecordAddr(&dns.A{A: net.ParseIP("100.64.0.7")})
+		require.True(t, ok)
+		assert.True(t, addr.Is4())
+		assert.Equal(t, netip.MustParseAddr("100.64.0.7"), addr)
+	})
+
+	t.Run("AAAA record yields v6", func(t *testing.T) {
+		addr, ok := extractRecordAddr(&dns.AAAA{AAAA: net.ParseIP("fd00::1")})
+		require.True(t, ok)
+		assert.Equal(t, netip.MustParseAddr("fd00::1"), addr)
+	})
+
+	t.Run("A record without address", func(t *testing.T) {
+		_, ok := extractRecordAddr(&dns.A{})
+		assert.False(t, ok)
+	})
+
+	t.Run("non-address record", func(t *testing.T) {
+		_, ok := extractRecordAddr(&dns.CNAME{Target: "target.netbird.cloud."})
+		assert.False(t, ok)
+	})
+}

client/internal/dns/mock_server.go

@@ -8,6 +8,7 @@ import (
 	"github.com/miekg/dns"
 
 	dnsconfig "github.com/netbirdio/netbird/client/internal/dns/config"
+	"github.com/netbirdio/netbird/client/internal/dns/local"
 	nbdns "github.com/netbirdio/netbird/dns"
 	"github.com/netbirdio/netbird/route"
 	"github.com/netbirdio/netbird/shared/management/domain"
@@ -92,6 +93,11 @@ func (m *MockServer) SetFirewall(Firewall) {
 	// Mock implementation - no-op
 }
 
+// SetPeerActivator mock implementation of SetPeerActivator from Server interface
+func (m *MockServer) SetPeerActivator(local.PeerActivator) {
+	// Mock implementation - no-op
+}
+
 // BeginBatch mock implementation of BeginBatch from Server interface
 func (m *MockServer) BeginBatch() {
 	// Mock implementation - no-op

client/internal/dns/resutil/resolve.go

@@ -19,6 +19,34 @@ import (
 // uses when a resolved host has no addresses of the requested family.
 const errNoSuitableAddress = "no suitable address found"
 
+// Extended DNS Error info codes NetBird emits so a client can see why an answer
+// looks the way it does without reading this peer's logs. They live in the RFC
+// 8914 Private Use range (49152-65535) and are registered here, in one place,
+// because nothing else guarantees two NetBird components pick distinct codes.
+const (
+	// EDENetbirdUpstreamTimeout: a DNS forwarder's upstream did not answer.
+	EDENetbirdUpstreamTimeout uint16 = 49152
+	// EDENetbirdUpstreamFailure: a DNS forwarder's upstream failed.
+	EDENetbirdUpstreamFailure uint16 = 49153
+	// EDENetbirdSoftenedNegative: the empty answer is ours, not the answering
+	// resolver's. A handler that owns the name deferred this query type, and the
+	// negative verdict that came back was downgraded so it cannot poison the
+	// name.
+	EDENetbirdSoftenedNegative uint16 = 49154
+)
+
+// AttachEDE adds an Extended DNS Error (RFC 8914) option to a message, creating
+// the OPT pseudo-record if it has none. Callers must only use it toward a client
+// that advertised EDNS0: per RFC 6891 an OPT must not appear otherwise.
+func AttachEDE(msg *dns.Msg, code uint16, text string) {
+	opt := msg.IsEdns0()
+	if opt == nil {
+		msg.SetEdns0(dns.DefaultMsgSize, false)
+		opt = msg.IsEdns0()
+	}
+	opt.Option = append(opt.Option, &dns.EDNS0_EDE{InfoCode: code, ExtraText: text})
+}
+
 // GenerateRequestID creates a random 8-character hex string for request tracing.
 func GenerateRequestID() string {
 	bytes := make([]byte, 4)
@@ -78,10 +106,38 @@ type resolver interface {
 	LookupNetIP(ctx context.Context, network, host string) ([]netip.Addr, error)
 }
 
+// MetaKey names an annotation a handler attaches to a request to explain the
+// response the chain ends up writing. The set is closed and rendered onto a
+// single log line, so an unrecognized key is a typo inventing a field rather
+// than a new fact.
+type MetaKey string
+
+const (
+	// MetaKeyProtocol: the transport a query arrived on.
+	MetaKeyProtocol MetaKey = "protocol"
+	// MetaKeyUpstream: the upstream that answered.
+	MetaKeyUpstream MetaKey = "upstream"
+	// MetaKeyUpstreamProtocol: the transport used toward that upstream.
+	MetaKeyUpstreamProtocol MetaKey = "upstream_protocol"
+	// MetaKeyPeer: the routing peer whose forwarder answered.
+	MetaKeyPeer MetaKey = "peer"
+	// MetaKeyEDE: an Extended DNS Error carried by the answer.
+	MetaKeyEDE MetaKey = "ede"
+	// MetaKeyTruncated: the answer did not fit and was truncated.
+	MetaKeyTruncated MetaKey = "truncated"
+	// MetaKeySoftened: a negative verdict was rewritten so it cannot poison a
+	// name served locally.
+	MetaKeySoftened MetaKey = "softened"
+	// MetaKeyDeferredBy: the handler that owned the name and stepped aside.
+	MetaKeyDeferredBy MetaKey = "deferred_by"
+	// MetaKeyDeferredReason: why it stepped aside.
+	MetaKeyDeferredReason MetaKey = "deferred_reason"
+)
+
 // chainedWriter is implemented by ResponseWriters that carry request metadata
 type chainedWriter interface {
 	RequestID() string
-	SetMeta(key, value string)
+	SetMeta(key MetaKey, value string)
 }
 
 // GetRequestID extracts a request ID from the ResponseWriter if available,
@@ -96,12 +152,30 @@ func GetRequestID(w dns.ResponseWriter) string {
 }
 
 // SetMeta sets metadata on the ResponseWriter if it supports it.
-func SetMeta(w dns.ResponseWriter, key, value string) {
+func SetMeta(w dns.ResponseWriter, key MetaKey, value string) {
 	if cw, ok := w.(chainedWriter); ok {
 		cw.SetMeta(key, value)
 	}
 }
 
+// softNegativeRequester is implemented by chain writers that can soften the
+// negative verdict of the handlers a deferring handler falls through to.
+type softNegativeRequester interface {
+	RequestSoftNegative()
+}
+
+// RequestSoftNegative asks the handler chain to downgrade an NXDOMAIN from the
+// handlers that run after the caller defers. A handler that owns a name but
+// cannot answer one query type for it needs this: the resolvers it falls
+// through to cannot prove the name absent, and an NXDOMAIN from them is cached
+// for the name and every type under it (RFC 2308, RFC 8020), taking the
+// addresses the handler does serve down with it.
+func RequestSoftNegative(w dns.ResponseWriter) {
+	if sn, ok := w.(softNegativeRequester); ok {
+		sn.RequestSoftNegative()
+	}
+}
+
 // LookupResult contains the result of an external DNS lookup
 type LookupResult struct {
 	IPs   []netip.Addr
@@ -199,11 +273,27 @@ type RecordResolver interface {
 	LookupAddr(ctx context.Context, addr string) ([]string, error)
 }
 
+// SupportedRecordQtype reports whether LookupRecords can resolve qtype. The
+// set is bounded by the net.Resolver API, which exposes no way to query an
+// arbitrary record type, and going around it with a raw exchange would mean
+// picking nameservers ourselves instead of resolving the way the host does.
+//
+// Both ends read this: a DNS forwarder answers these types for the domains it
+// routes, and a client uses it to tell which types are worth forwarding to a
+// peer at all.
+func SupportedRecordQtype(qtype uint16) bool {
+	switch qtype {
+	case dns.TypeMX, dns.TypeTXT, dns.TypeNS, dns.TypeSRV, dns.TypeCNAME, dns.TypePTR:
+		return true
+	default:
+		return false
+	}
+}
+
 // LookupRecords resolves a non-address DNS record type through the host
-// resolver and returns the resource records and the DNS rcode. Types the host
-// resolver cannot answer (anything not covered by the net.Resolver Lookup*
-// methods) yield NODATA so that a routed name is never poisoned with NXDOMAIN
-// for an unsupported type.
+// resolver and returns the resource records and the DNS rcode. Types outside
+// SupportedRecordQtype yield NODATA so that a routed name is never poisoned
+// with NXDOMAIN for a type we cannot look up.
 func LookupRecords(ctx context.Context, r RecordResolver, name string, qtype uint16, ttl uint32) ([]dns.RR, int) {
 	fqdn := dns.Fqdn(name)
 

client/internal/dns/server.go

@@ -82,6 +82,7 @@ type Server interface {
 	PopulateManagementDomain(mgmtURL *url.URL) error
 	SetRouteSources(selected, active func() route.HAMap)
 	SetFirewall(Firewall)
+	SetPeerActivator(local.PeerActivator)
 }
 
 type nsGroupsByDomain struct {
@@ -491,6 +492,13 @@ func (s *DefaultServer) SetFirewall(fw Firewall) {
 	}
 }
 
+// SetPeerActivator wires the DNS-time lazy-connection warm-up on the local
+// resolver. Injected after the connection manager exists (it does not at
+// DNS-server construction time). Pass nil to disable.
+func (s *DefaultServer) SetPeerActivator(a local.PeerActivator) {
+	s.localResolver.SetPeerActivator(a)
+}
+
 // Stop stops the server
 func (s *DefaultServer) Stop() {
 	s.ctxCancel()
@@ -1435,11 +1443,11 @@ type localPeerConnectivity struct {
 
 // IsConnectedByIP looks the IP up in the peerstore and surfaces both
 // the known and connected bits. Used by Resolver.filterDisconnectedPeerAnswers.
-func (l localPeerConnectivity) IsConnectedByIP(ip string) (known, connected bool) {
+func (l localPeerConnectivity) IsConnectedByIP(ip netip.Addr) (known, connected bool) {
 	if l.status == nil {
 		return false, false
 	}
-	state, ok := l.status.PeerStateByIP(ip)
+	state, ok := l.status.PeerStateByIP(ip.String())
 	if !ok {
 		return false, false
 	}

client/internal/dns/upstream.go

@@ -295,7 +295,7 @@ func (u *upstreamResolverBase) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 	if addr := w.RemoteAddr(); addr != nil {
 		network := addr.Network()
 		ctx = contextWithDNSProtocol(ctx, network)
-		resutil.SetMeta(w, "protocol", network)
+		resutil.SetMeta(w, resutil.MetaKeyProtocol, network)
 	}
 
 	ok, failures := u.tryUpstreamServers(ctx, w, r, logger)
@@ -331,7 +331,7 @@ func (u *upstreamResolverBase) tryOnlyRace(ctx context.Context, w dns.ResponseWr
 		return false, res.failures
 	}
 	if res.ede != "" {
-		resutil.SetMeta(w, "ede", res.ede)
+		resutil.SetMeta(w, resutil.MetaKeyEDE, res.ede)
 	}
 	u.writeSuccessResponse(w, res.msg, res.upstream, r.Question[0].Name, res.protocol, logger)
 	return true, res.failures
@@ -361,7 +361,7 @@ func (u *upstreamResolverBase) raceAll(ctx context.Context, w dns.ResponseWriter
 			failures = append(failures, res.failures...)
 			if res.msg != nil {
 				if res.ede != "" {
-					resutil.SetMeta(w, "ede", res.ede)
+					resutil.SetMeta(w, resutil.MetaKeyEDE, res.ede)
 				}
 				u.writeSuccessResponse(w, res.msg, res.upstream, r.Question[0].Name, res.protocol, logger)
 				return true, failures
@@ -550,9 +550,9 @@ func (u *upstreamResolverBase) debugUpstreamTimeout(upstream netip.AddrPort) str
 }
 
 func (u *upstreamResolverBase) writeSuccessResponse(w dns.ResponseWriter, rm *dns.Msg, upstream netip.AddrPort, domain string, proto string, logger *log.Entry) {
-	resutil.SetMeta(w, "upstream", upstream.String())
+	resutil.SetMeta(w, resutil.MetaKeyUpstream, upstream.String())
 	if proto != "" {
-		resutil.SetMeta(w, "upstream_protocol", proto)
+		resutil.SetMeta(w, resutil.MetaKeyUpstreamProtocol, proto)
 	}
 
 	// Clear Zero bit from external responses to prevent upstream servers from

client/internal/dns_peer_activator.go

@@ -0,0 +1,76 @@
+package internal
+
+import (
+	"context"
+	"net/netip"
+	"time"
+
+	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/client/internal/peerstore"
+)
+
+const dnsActivationPollInterval = 50 * time.Millisecond
+
+// dnsPeerActivator wakes lazy-connection peers from the DNS resolution path. It
+// implements dns/local.PeerActivator. DNS queries run on their own goroutines,
+// so it only touches state that is safe for concurrent use — ConnMgr.ActivatePeer,
+// peerstore.Store and peer.Status — and never takes the engine's syncMsgMux,
+// keeping DNS resolution from contending with network-map processing.
+type dnsPeerActivator struct {
+	connMgr   *ConnMgr
+	peerStore *peerstore.Store
+	status    *peer.Status
+	// ctx is the engine's long-lived context. The connection dial is tied to it
+	// (not the per-query DNS wait budget) so a handshake that outlasts the wait
+	// still completes in the background rather than being cancelled at the deadline.
+	ctx context.Context
+}
+
+// ActivatePeersByIP triggers wake-up for the peer(s) owning addrs and waits
+// until one is connected or ctx (the per-query DNS wait budget) expires.
+// Activation itself is tied to the engine's long-lived context so the dial
+// survives a wait that times out. Unknown or already-connected addresses are
+// skipped, so the steady-state (warm) path adds no latency.
+func (a *dnsPeerActivator) ActivatePeersByIP(ctx context.Context, addrs []netip.Addr) {
+	if a == nil || a.connMgr == nil {
+		return
+	}
+
+	var pending []string
+	for _, addr := range addrs {
+		ip := addr.String()
+		st, ok := a.status.PeerStateByIP(ip)
+		if !ok || st.ConnStatus == peer.StatusConnected {
+			continue
+		}
+		conn, ok := a.peerStore.PeerConn(st.PubKey)
+		if !ok {
+			continue
+		}
+		a.connMgr.ActivatePeer(a.ctx, conn)
+		pending = append(pending, ip)
+	}
+
+	if len(pending) == 0 {
+		return
+	}
+	a.waitConnected(ctx, pending)
+}
+
+// waitConnected blocks until any of ips reports a connected peer or ctx expires.
+func (a *dnsPeerActivator) waitConnected(ctx context.Context, ips []string) {
+	ticker := time.NewTicker(dnsActivationPollInterval)
+	defer ticker.Stop()
+	for {
+		for _, ip := range ips {
+			if st, ok := a.status.PeerStateByIP(ip); ok && st.ConnStatus == peer.StatusConnected {
+				return
+			}
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+		}
+	}
+}

client/internal/dns_peer_activator_test.go

@@ -0,0 +1,129 @@
+package internal
+
+import (
+	"context"
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/client/internal/peerstore"
+)
+
+func newTestPeerConn(t *testing.T, key string) *peer.Conn {
+	t.Helper()
+	conn, err := peer.NewConn(peer.ConnConfig{
+		Key:      key,
+		LocalKey: "local",
+		WgConfig: peer.WgConfig{
+			AllowedIps: []netip.Prefix{netip.MustParsePrefix("100.64.0.1/32")},
+		},
+	}, peer.ServiceDependencies{})
+	require.NoError(t, err)
+	return conn
+}
+
+func newTestDNSPeerActivator(t *testing.T) (*dnsPeerActivator, *peer.Status, *peerstore.Store) {
+	t.Helper()
+	status := peer.NewRecorder("https://mgm")
+	store := peerstore.NewConnStore()
+	// ConnMgr without Start: the lazy manager is nil, so ActivatePeer is a
+	// no-op — these tests exercise the activator's skip/wait logic.
+	connMgr := NewConnMgr(&EngineConfig{}, status, store, nil)
+	return &dnsPeerActivator{
+		connMgr:   connMgr,
+		peerStore: store,
+		status:    status,
+		ctx:       context.Background(),
+	}, status, store
+}
+
+func TestDNSPeerActivator_NilSafe(t *testing.T) {
+	var a *dnsPeerActivator
+	a.ActivatePeersByIP(context.Background(), []netip.Addr{netip.MustParseAddr("100.64.0.1")})
+}
+
+// TestDNSPeerActivator_SkipsUnknownAndConnectedPeers verifies the steady-state
+// (warm) path adds no latency: already-connected and unknown addresses never
+// enter the wait loop.
+func TestDNSPeerActivator_SkipsUnknownAndConnectedPeers(t *testing.T) {
+	a, status, store := newTestDNSPeerActivator(t)
+
+	require.NoError(t, status.AddPeer("peerA", "a.netbird.cloud", "100.64.0.1", "fd00::1"))
+	require.NoError(t, status.UpdatePeerState(peer.State{PubKey: "peerA", ConnStatus: peer.StatusConnected}))
+	store.AddPeerConn("peerA", newTestPeerConn(t, "peerA"))
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	start := time.Now()
+	a.ActivatePeersByIP(ctx, []netip.Addr{
+		netip.MustParseAddr("100.64.0.1"),  // known, connected -> skipped
+		netip.MustParseAddr("fd00::1"),     // known via IPv6, connected -> skipped
+		netip.MustParseAddr("100.64.0.99"), // unknown -> skipped
+	})
+	require.Less(t, time.Since(start), time.Second, "no pending peer must mean no wait")
+}
+
+// TestDNSPeerActivator_WaitsForPendingPeerToConnect verifies the wait loop
+// returns as soon as a pending peer reports connected, well before the
+// per-query budget expires.
+func TestDNSPeerActivator_WaitsForPendingPeerToConnect(t *testing.T) {
+	a, status, store := newTestDNSPeerActivator(t)
+
+	require.NoError(t, status.AddPeer("peerA", "a.netbird.cloud", "100.64.0.1", ""))
+	store.AddPeerConn("peerA", newTestPeerConn(t, "peerA"))
+
+	go func() {
+		time.Sleep(150 * time.Millisecond)
+		_ = status.UpdatePeerState(peer.State{PubKey: "peerA", ConnStatus: peer.StatusConnected})
+	}()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	start := time.Now()
+	a.ActivatePeersByIP(ctx, []netip.Addr{netip.MustParseAddr("100.64.0.1")})
+	elapsed := time.Since(start)
+
+	require.GreaterOrEqual(t, elapsed, 100*time.Millisecond, "must wait for the pending peer")
+	require.Less(t, elapsed, 5*time.Second, "must return on connect, not at the deadline")
+}
+
+// TestDNSPeerActivator_ReturnsAtBudgetWhenPeerStaysIdle verifies a peer that
+// never connects releases the DNS response at the per-query budget instead of
+// blocking it indefinitely.
+func TestDNSPeerActivator_ReturnsAtBudgetWhenPeerStaysIdle(t *testing.T) {
+	a, status, store := newTestDNSPeerActivator(t)
+
+	require.NoError(t, status.AddPeer("peerA", "a.netbird.cloud", "100.64.0.1", ""))
+	store.AddPeerConn("peerA", newTestPeerConn(t, "peerA"))
+
+	ctx, cancel := context.WithTimeout(context.Background(), 300*time.Millisecond)
+	defer cancel()
+
+	start := time.Now()
+	a.ActivatePeersByIP(ctx, []netip.Addr{netip.MustParseAddr("100.64.0.1")})
+	elapsed := time.Since(start)
+
+	require.GreaterOrEqual(t, elapsed, 250*time.Millisecond, "must wait out the budget for a pending peer")
+	require.Less(t, elapsed, 5*time.Second, "must not block past the budget")
+}
+
+// TestDNSPeerActivator_NoWaitWithoutPeerConn verifies a known-but-idle peer
+// with no connection object in the store is not waited on: there is nothing to
+// activate, so waiting could only ever time out.
+func TestDNSPeerActivator_NoWaitWithoutPeerConn(t *testing.T) {
+	a, status, _ := newTestDNSPeerActivator(t)
+
+	require.NoError(t, status.AddPeer("peerA", "a.netbird.cloud", "100.64.0.1", ""))
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	start := time.Now()
+	a.ActivatePeersByIP(ctx, []netip.Addr{netip.MustParseAddr("100.64.0.1")})
+	require.Less(t, time.Since(start), time.Second, "peer without a conn must not be waited on")
+}

client/internal/dnsfwd/forwarder.go

@@ -26,15 +26,6 @@ import (
 const errResolveFailed = "failed to resolve query for domain=%s: %v"
 const upstreamTimeout = 15 * time.Second
 
-// EDE info codes the forwarder emits on upstream failures so the querying
-// client can see the reason without inspecting this peer's logs. They live in
-// the RFC 8914 Private Use range (49152-65535); the Go resolver never exposes a
-// real upstream EDE here, so these cannot collide with a genuine code.
-const (
-	edeNetbirdUpstreamTimeout uint16 = 49152
-	edeNetbirdUpstreamFailure uint16 = 49153
-)
-
 type resolver interface {
 	LookupNetIP(ctx context.Context, network, host string) ([]netip.Addr, error)
 	LookupMX(ctx context.Context, name string) ([]*net.MX, error)
@@ -216,6 +207,9 @@ func (f *DNSForwarder) handleDNSQuery(logger *log.Entry, w dns.ResponseWriter, q
 		qname, dns.TypeToString[question.Qtype], dns.ClassToString[question.Qclass])
 
 	resp := query.SetReply(query)
+	// Every answer here comes from a recursive lookup on this peer. SetReply
+	// leaves RA unset, which reads to a client as a server that cannot recurse.
+	resp.RecursionAvailable = true
 
 	mostSpecificResId, matchingEntries := f.getMatchingEntries(strings.TrimSuffix(qname, "."))
 	if mostSpecificResId == "" {
@@ -229,20 +223,22 @@ func (f *DNSForwarder) handleDNSQuery(logger *log.Entry, w dns.ResponseWriter, q
 
 	reqHasEdns := query.IsEdns0() != nil
 
-	switch question.Qtype {
-	case dns.TypeA, dns.TypeAAAA:
+	switch {
+	case question.Qtype == dns.TypeA || question.Qtype == dns.TypeAAAA:
 		f.handleAddressQuery(ctx, logger, w, resp, mostSpecificResId, matchingEntries, reqHasEdns, startTime)
-	case dns.TypeMX, dns.TypeTXT, dns.TypeNS, dns.TypeSRV, dns.TypeCNAME, dns.TypePTR:
+	case resutil.SupportedRecordQtype(question.Qtype):
 		f.handleRecordQuery(ctx, logger, w, resp, startTime)
 	default:
 		// The domain is routed here, so any other type is answered NODATA
 		// (NOERROR, empty answer) rather than falling back to a resolver that
 		// would poison the name with NXDOMAIN. The Extended DNS Error lets a
 		// client tell this capability-driven NODATA apart from an
-		// authoritative one. The OPT pseudo-record must not appear unless the
-		// query advertised EDNS0.
+		// authoritative one; a current client knows the type is unsupported and
+		// resolves it through its own handler chain instead of asking at all.
+		// The OPT pseudo-record must not appear unless the query advertised
+		// EDNS0.
 		if reqHasEdns {
-			attachEDE(resp, dns.ExtendedErrorCodeNotSupported, "netbird forwarder: unsupported query type")
+			resutil.AttachEDE(resp, dns.ExtendedErrorCodeNotSupported, "netbird forwarder: unsupported query type")
 		}
 		f.writeResponse(logger, w, resp, qname, startTime)
 	}
@@ -441,7 +437,7 @@ func (f *DNSForwarder) handleDNSError(
 	}
 
 	if reqHasEdns {
-		attachEDE(resp, edeCodeFor(dnsErr), edeText(dnsErr))
+		resutil.AttachEDE(resp, edeCodeFor(dnsErr), edeText(dnsErr))
 	}
 
 	f.writeResponse(logger, w, resp, domain, startTime)
@@ -488,9 +484,9 @@ func (f *DNSForwarder) getMatchingEntries(domain string) (route.ResID, []*Forwar
 // edeCodeFor maps an upstream lookup error to the NetBird EDE info code.
 func edeCodeFor(dnsErr *net.DNSError) uint16 {
 	if dnsErr != nil && dnsErr.IsTimeout {
-		return edeNetbirdUpstreamTimeout
+		return resutil.EDENetbirdUpstreamTimeout
 	}
-	return edeNetbirdUpstreamFailure
+	return resutil.EDENetbirdUpstreamFailure
 }
 
 // edeText builds the EDE extra-text describing the class of upstream failure.
@@ -503,14 +499,3 @@ func edeText(dnsErr *net.DNSError) string {
 	}
 	return "netbird forwarder: upstream failure"
 }
-
-// attachEDE adds an Extended DNS Error (RFC 8914) option to the response,
-// creating the OPT pseudo-record if the response does not already carry one.
-func attachEDE(resp *dns.Msg, code uint16, text string) {
-	opt := resp.IsEdns0()
-	if opt == nil {
-		resp.SetEdns0(dns.DefaultMsgSize, false)
-		opt = resp.IsEdns0()
-	}
-	opt.Option = append(opt.Option, &dns.EDNS0_EDE{InfoCode: code, ExtraText: text})
-}

client/internal/dnsfwd/forwarder_test.go

@@ -649,6 +649,39 @@ func TestDNSForwarder_ResponseCodes(t *testing.T) {
 	}
 }
 
+// TestDNSForwarder_RecursionAvailable covers the RA bit. Every answer this
+// forwarder produces comes from a recursive lookup on the routing peer, but
+// dns.Msg.SetReply leaves RA unset, so clients report "recursion not available"
+// and some stub resolvers treat the server as unable to serve the query.
+func TestDNSForwarder_RecursionAvailable(t *testing.T) {
+	t.Run("record answer", func(t *testing.T) {
+		mockResolver := &MockResolver{}
+		forwarder := newRecordTestForwarder(t, mockResolver, "example.com")
+
+		mockResolver.On("LookupMX", mock.Anything, "example.com.").
+			Return([]*net.MX{{Host: "mail.example.com.", Pref: 10}}, nil).Once()
+
+		resp := runRecordQuery(t, forwarder, "example.com", dns.TypeMX)
+		assert.True(t, resp.RecursionAvailable, "the forwarder resolves recursively")
+	})
+
+	t.Run("unsupported type NODATA", func(t *testing.T) {
+		forwarder := newRecordTestForwarder(t, &MockResolver{}, "example.com")
+
+		resp := runRecordQuery(t, forwarder, "example.com", dns.TypeCAA)
+		require.Equal(t, dns.RcodeSuccess, resp.Rcode)
+		assert.True(t, resp.RecursionAvailable, "RA describes the server, not the query type")
+	})
+
+	t.Run("unauthorized domain", func(t *testing.T) {
+		forwarder := newRecordTestForwarder(t, &MockResolver{}, "example.com")
+
+		resp := runRecordQuery(t, forwarder, "other.com", dns.TypeMX)
+		require.Equal(t, dns.RcodeRefused, resp.Rcode)
+		assert.True(t, resp.RecursionAvailable, "RA describes the server, not the verdict")
+	})
+}
+
 func hasEDE(m *dns.Msg, code uint16) bool {
 	opt := m.IsEdns0()
 	if opt == nil {
@@ -859,7 +892,7 @@ func TestDNSForwarder_UpstreamFailureEDE(t *testing.T) {
 			lookupErr:   &net.DNSError{Err: "i/o timeout", Server: "10.0.0.53:53", IsTimeout: true},
 			reqEdns:     true,
 			wantEDE:     true,
-			wantCode:    edeNetbirdUpstreamTimeout,
+			wantCode:    resutil.EDENetbirdUpstreamTimeout,
 			wantTextHas: "netbird forwarder: upstream timeout",
 		},
 		{
@@ -867,7 +900,7 @@ func TestDNSForwarder_UpstreamFailureEDE(t *testing.T) {
 			lookupErr:   &net.DNSError{Err: "server misbehaving", Server: "10.0.0.53:53"},
 			reqEdns:     true,
 			wantEDE:     true,
-			wantCode:    edeNetbirdUpstreamFailure,
+			wantCode:    resutil.EDENetbirdUpstreamFailure,
 			wantTextHas: "netbird forwarder: upstream failure",
 		},
 		{

client/internal/ebpf/ebpf/src/dns_fwd.c

@@ -52,11 +52,14 @@ int xdp_dns_fwd(struct iphdr  *ip, struct udphdr *udp) {
 
     if (udp->dest == GENERAL_DNS_PORT && ip->daddr == dns_ip) {
         udp->dest = dns_port;
+        // Clear the now-stale checksum; zero means "not computed" for IPv4.
+        udp->check = 0;
         return XDP_PASS;
     }
 
     if (udp->source == dns_port && ip->saddr == dns_ip) {
         udp->source = GENERAL_DNS_PORT;
+        udp->check = 0;
         return XDP_PASS;
     }
 

client/internal/ebpf/ebpf/src/wg_proxy.c

@@ -50,5 +50,11 @@ int xdp_wg_proxy(struct iphdr  *ip, struct udphdr *udp) {
     __be16 new_dst_port = htons(proxy_port);
     udp->dest = new_dst_port;
     udp->source = new_src_port;
+
+    // The ports are covered by the UDP checksum. This is an IPv4 loopback hop
+    // and the payload is already integrity-protected, so clear the checksum (a
+    // zero UDP checksum means "not computed" for IPv4) rather than leave a
+    // stale value the kernel would drop as UDP_CSUM.
+    udp->check = 0;
 	return XDP_PASS;
 }

client/internal/engine.go

@@ -64,7 +64,10 @@ import (
 	"github.com/netbirdio/netbird/route"
 	mgm "github.com/netbirdio/netbird/shared/management/client"
 	"github.com/netbirdio/netbird/shared/management/domain"
+	sharedgrpc "github.com/netbirdio/netbird/shared/management/grpc"
+	nbnetworkmap "github.com/netbirdio/netbird/shared/management/networkmap"
 	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+	types "github.com/netbirdio/netbird/shared/management/types"
 	"github.com/netbirdio/netbird/shared/netiputil"
 	auth "github.com/netbirdio/netbird/shared/relay/auth/hmac"
 	relayClient "github.com/netbirdio/netbird/shared/relay/client"
@@ -147,6 +150,7 @@ type EngineConfig struct {
 	BlockLANAccess      bool
 	BlockInbound        bool
 	DisableIPv6         bool
+	SyncMessageVersion  *int
 
 	// LazyConnection is the MDM-sourced lazy-connection override; StateUnset defers to
 	// the env var and management feature flag.
@@ -220,6 +224,13 @@ type Engine struct {
 	// networkSerial is the latest CurrentSerial (state ID) of the network sent by the Management service
 	networkSerial uint64
 
+	// latestComponents is the most-recent NetworkMapComponents decoded from
+	// a NetworkMapEnvelope (capability=3 peers only). Held alongside the
+	// NetworkMap that Calculate() produced from it so future incremental
+	// updates have a base to apply changes against. nil for legacy-format
+	// peers. Guarded by syncMsgMux.
+	latestComponents *types.NetworkMapComponents
+
 	networkMonitor *networkmonitor.NetworkMonitor
 
 	sshServer sshServer
@@ -652,8 +663,24 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 	iceCfg := e.createICEConfig()
 
 	e.connMgr = NewConnMgr(e.config, e.statusRecorder, e.peerStore, wgIface)
+	e.connMgr.SetRoutedIPsReconciler(func(peerKey string) error {
+		if e.routeManager == nil {
+			return nil
+		}
+		return e.routeManager.ReconcilePeerAllowedIPs(peerKey)
+	})
 	e.connMgr.Start(e.ctx)
 
+	// Wire DNS-time lazy-connection warm-up now that the connection manager
+	// exists (it does not at DNS-server construction time). A DNS answer that
+	// points at an idle peer then wakes it before the client's first request.
+	e.dnsServer.SetPeerActivator(&dnsPeerActivator{
+		connMgr:   e.connMgr,
+		peerStore: e.peerStore,
+		status:    e.statusRecorder,
+		ctx:       e.ctx,
+	})
+
 	e.srWatcher = guard.NewSRWatcher(e.signal, e.relayManager, e.mobileDep.IFaceDiscover, iceCfg)
 	e.srWatcher.Start(peer.IsForceRelayed())
 
@@ -963,8 +990,12 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 
 	e.ApplySessionDeadline(update.GetSessionExpiresAt())
 
-	if update.NetworkMap != nil && update.NetworkMap.PeerConfig != nil {
-		e.handleAutoUpdateVersion(update.NetworkMap.PeerConfig.AutoUpdate)
+	// Envelope sync responses carry PeerConfig at the top level; legacy
+	// NetworkMap syncs carry it under NetworkMap.PeerConfig.
+	if pc := update.GetPeerConfig(); pc != nil {
+		e.handleAutoUpdateVersion(pc.GetAutoUpdate())
+	} else if nm := update.GetNetworkMap(); nm != nil && nm.GetPeerConfig() != nil {
+		e.handleAutoUpdateVersion(nm.GetPeerConfig().GetAutoUpdate())
 	}
 
 	done := e.phase("netbird_config")
@@ -974,12 +1005,47 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 		return err
 	}
 
+	// Decode the network map from either the components envelope or the
+	// legacy proto.NetworkMap before the posture-check gating below, so the
+	// "is there a network map" decision covers both wire shapes.
+	var (
+		nm         *mgmProto.NetworkMap
+		components *types.NetworkMapComponents
+	)
+	if version := update.GetVersion(); version == int32(sharedgrpc.ComponentNetworkMap) {
+		// Components-format peer: decode the envelope back to typed
+		// components, run Calculate() locally, and convert to the wire
+		// NetworkMap shape the rest of the engine consumes. Components are
+		// retained so future incremental updates can apply deltas instead
+		// of doing a full reconstruction.
+		envelope := update.GetNetworkMapEnvelope()
+		if envelope == nil {
+			return fmt.Errorf("received a SyncReponse indicating use of components network map, but components are missing")
+		}
+
+		localKey := e.config.WgPrivateKey.PublicKey().String()
+		dnsName := ""
+		if pc := update.GetPeerConfig(); pc != nil {
+			// PeerConfig.Fqdn = "<dns_label>.<dns_domain>" — extract the
+			// shared domain by stripping the peer's own label prefix. Falls
+			// back to empty if the FQDN doesn't have the expected shape.
+			dnsName = extractDNSDomainFromFQDN(pc.GetFqdn())
+		}
+		result, err := nbnetworkmap.EnvelopeToNetworkMap(e.ctx, envelope, localKey, dnsName)
+		if err != nil {
+			return fmt.Errorf("decode network map envelope: %w", err)
+		}
+		nm = result.NetworkMap
+		components = result.Components
+	} else {
+		nm = update.GetNetworkMap()
+	}
+
 	// Posture checks are bound to the network map presence:
 	//   NetworkMap != nil, checks present -> apply the received checks
 	//   NetworkMap != nil, checks nil     -> posture checks were removed, clear them
 	//   NetworkMap == nil                 -> config-only update (e.g. relay token rotation),
 	//                                        leave the previously applied checks untouched
-	nm := update.GetNetworkMap()
 	if nm == nil {
 		return nil
 	}
@@ -992,6 +1058,14 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 	}
 
 	done = e.phase("persist")
+	// Only retain the components view when the server sent the envelope
+	// path. A legacy proto.NetworkMap means components == nil; writing it
+	// here would clobber a previously-cached snapshot, breaking the
+	// incremental-delta base on a future envelope sync.
+	if components != nil {
+		e.latestComponents = components
+	}
+
 	e.persistSyncResponse(update)
 	done()
 
@@ -1005,6 +1079,19 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 	return nil
 }
 
+// extractDNSDomainFromFQDN returns the trailing dotted domain part of the
+// receiving peer's FQDN — the same value the management server fills as
+// dnsName when it builds the legacy NetworkMap. "peer42.netbird.cloud" →
+// "netbird.cloud". An empty string is returned for unrecognized formats.
+func extractDNSDomainFromFQDN(fqdn string) string {
+	for i := 0; i < len(fqdn); i++ {
+		if fqdn[i] == '.' && i+1 < len(fqdn) {
+			return fqdn[i+1:]
+		}
+	}
+	return ""
+}
+
 // updateNetbirdConfig applies the management-provided NetBird configuration:
 // STUN/TURN and relay servers, flow logging and DNS settings. A nil config is a no-op,
 // which is the case for sync updates carrying only a network map.
@@ -1164,6 +1251,7 @@ func (e *Engine) applyInfoFlags(info *system.Info) {
 		e.config.BlockLANAccess,
 		e.config.BlockInbound,
 		e.config.DisableIPv6,
+		e.config.SyncMessageVersion,
 		e.config.EnableSSHRoot,
 		e.config.EnableSSHSFTP,
 		e.config.EnableSSHLocalPortForwarding,
@@ -2032,6 +2120,7 @@ func (e *Engine) readInitialSettings() ([]*route.Route, *nbdns.Config, bool, err
 		e.config.BlockLANAccess,
 		e.config.BlockInbound,
 		e.config.DisableIPv6,
+		e.config.SyncMessageVersion,
 		e.config.EnableSSHRoot,
 		e.config.EnableSSHSFTP,
 		e.config.EnableSSHLocalPortForwarding,

client/internal/engine_session_deadline_test.go

@@ -75,4 +75,14 @@ func TestApplySessionDeadline_ThreeState(t *testing.T) {
 		require.True(t, e.statusRecorder.GetSessionExpiresAt().IsZero(),
 			"invalid timestamp must clear the deadline")
 	})
+
+	t.Run("recently expired timestamp stays visible as expired", func(t *testing.T) {
+		e := newEngine()
+		expired := time.Now().Add(-5 * time.Minute).UTC().Truncate(time.Second)
+
+		e.ApplySessionDeadline(timestamppb.New(expired))
+
+		require.True(t, e.statusRecorder.GetSessionExpiresAt().Equal(expired),
+			"recently-expired deadline must stay on the recorder so consumers render it as expired")
+	})
 }

client/internal/lazyconn/manager/manager.go

@@ -29,6 +29,11 @@ type managedPeer struct {
 
 type Config struct {
 	InactivityThreshold *time.Duration
+	// ReconcileAllowedIPs re-applies a peer's routed allowed IPs after its wake endpoint is
+	// armed. The activity listener creates the wake peer with the overlay /32 only; without the
+	// routed prefixes WireGuard would not steer subnet-bound traffic to the wake endpoint, so an
+	// idle routing peer could never be woken by that traffic. Optional; nil disables the reconcile.
+	ReconcileAllowedIPs func(peerKey string) error
 }
 
 // Manager manages lazy connections
@@ -56,6 +61,9 @@ type Manager struct {
 	peerToHAGroups map[string][]route.HAUniqueID // peer ID -> HA groups they belong to
 	haGroupToPeers map[route.HAUniqueID][]string // HA group -> peer IDs in the group
 	routesMu       sync.RWMutex
+
+	// reconcileAllowedIPs re-applies a peer's routed allowed IPs after its wake endpoint is armed.
+	reconcileAllowedIPs func(peerKey string) error
 }
 
 // NewManager creates a new lazy connection manager
@@ -73,6 +81,7 @@ func NewManager(config Config, engineCtx context.Context, peerStore *peerstore.S
 		activityManager:      activity.NewManager(wgIface),
 		peerToHAGroups:       make(map[string][]route.HAUniqueID),
 		haGroupToPeers:       make(map[route.HAUniqueID][]string),
+		reconcileAllowedIPs:  config.ReconcileAllowedIPs,
 	}
 
 	if wgIface.IsUserspaceBind() {
@@ -201,7 +210,7 @@ func (m *Manager) AddPeer(peerCfg lazyconn.PeerConfig) (bool, error) {
 		return false, nil
 	}
 
-	if err := m.activityManager.MonitorPeerActivity(peerCfg); err != nil {
+	if err := m.armActivityListener(peerCfg); err != nil {
 		return false, err
 	}
 
@@ -288,7 +297,7 @@ func (m *Manager) DeactivatePeer(peerID peerid.ConnID) {
 
 	m.inactivityManager.RemovePeer(mp.peerCfg.PublicKey)
 
-	if err := m.activityManager.MonitorPeerActivity(*mp.peerCfg); err != nil {
+	if err := m.armActivityListener(*mp.peerCfg); err != nil {
 		mp.peerCfg.Log.Errorf("failed to create activity monitor: %v", err)
 		return
 	}
@@ -465,6 +474,31 @@ func (m *Manager) close() {
 }
 
 // shouldDeferIdleForHA checks if peer should stay connected due to HA group requirements
+// armRoutedAllowedIPs re-applies the peer's routed allowed IPs onto its freshly armed wake
+// endpoint. The activity listener creates the wake peer with the overlay /32 only, so without
+// this the routed prefixes would be missing and traffic to a routed subnet could not wake the
+// idle routing peer. It is a no-op when no reconciler is configured.
+// armActivityListener (re)arms the peer's wake endpoint via the activity manager and then
+// re-applies its routed allowed IPs, so traffic to a routed subnet can wake an idle routing
+// peer. The routed prefixes must be re-applied after the wake endpoint exists because the
+// listener creates it with the overlay /32 only.
+func (m *Manager) armActivityListener(peerCfg lazyconn.PeerConfig) error {
+	if err := m.activityManager.MonitorPeerActivity(peerCfg); err != nil {
+		return err
+	}
+	m.armRoutedAllowedIPs(&peerCfg)
+	return nil
+}
+
+func (m *Manager) armRoutedAllowedIPs(peerCfg *lazyconn.PeerConfig) {
+	if m.reconcileAllowedIPs == nil {
+		return
+	}
+	if err := m.reconcileAllowedIPs(peerCfg.PublicKey); err != nil {
+		peerCfg.Log.Errorf("failed to reconcile routed allowed IPs on wake endpoint: %v", err)
+	}
+}
+
 func (m *Manager) shouldDeferIdleForHA(inactivePeers map[string]struct{}, peerID string) bool {
 	m.routesMu.RLock()
 	defer m.routesMu.RUnlock()
@@ -577,7 +611,7 @@ func (m *Manager) onPeerInactivityTimedOut(peerIDs map[string]struct{}) {
 
 		mp.peerCfg.Log.Infof("start activity monitor")
 
-		if err := m.activityManager.MonitorPeerActivity(*mp.peerCfg); err != nil {
+		if err := m.armActivityListener(*mp.peerCfg); err != nil {
 			mp.peerCfg.Log.Errorf("failed to create activity monitor: %v", err)
 			continue
 		}

client/internal/peer/status.go

@@ -813,19 +813,14 @@ func (d *Status) SetSessionExpiresAt(deadline time.Time) {
 }
 
 // GetSessionExpiresAt returns the most recently recorded SSO session deadline,
-// or the zero value when no deadline is tracked. A deadline that has already
-// slipped into the past reports as "none": once the session has expired it is
-// no longer a meaningful countdown, and the sessionwatch.Watcher does not
-// arm a timer at the deadline itself to clear it (only the two pre-expiry
-// warnings). Without this guard the UI would keep painting a stale
-// "expires in …" against a moment that has passed until the next login,
-// extend, or teardown rewrote the value.
+// or the zero value when no deadline is tracked. A deadline in the past is
+// returned as-is: it means the session has expired, and consumers (tray row,
+// CLI status) render it as "expired" rather than hiding it — masking it as
+// "none" would blank the UI at the exact moment it should say the session
+// ended.
 func (d *Status) GetSessionExpiresAt() time.Time {
 	d.mux.Lock()
 	defer d.mux.Unlock()
-	if !d.sessionExpiresAt.IsZero() && d.sessionExpiresAt.Before(time.Now()) {
-		return time.Time{}
-	}
 	return d.sessionExpiresAt
 }
 

client/internal/peer/worker_relay.go

@@ -54,15 +54,19 @@ func (w *WorkerRelay) OnNewOffer(remoteOfferAnswer *OfferAnswer) {
 	w.relaySupportedOnRemotePeer.Store(true)
 
 	// the relayManager will return with error in case if the connection has lost with relay server
-	_, _, err := w.relayManager.RelayInstanceAddress()
+	currentRelayAddress, _, err := w.relayManager.RelayInstanceAddress()
 	if err != nil {
 		w.log.Errorf("failed to handle new offer: %s", err)
 		return
 	}
 
-	preferForeign := !w.isController
-	remoteRelayServer := relayClient.RelayServer{Addr: remoteOfferAnswer.RelaySrvAddress, IP: remoteOfferAnswer.RelaySrvIP}
-	relayedConn, err := w.relayManager.OpenConn(w.peerCtx, remoteRelayServer, w.config.Key, preferForeign)
+	srv := w.preferredRelayServer(currentRelayAddress, remoteOfferAnswer.RelaySrvAddress)
+	var serverIP netip.Addr
+	if srv == remoteOfferAnswer.RelaySrvAddress {
+		serverIP = remoteOfferAnswer.RelaySrvIP
+	}
+
+	relayedConn, err := w.relayManager.OpenConn(w.peerCtx, srv, w.config.Key, serverIP)
 	if err != nil {
 		if errors.Is(err, relayClient.ErrConnAlreadyExists) {
 			w.log.Debugf("handled offer by reusing existing relay connection")
@@ -76,13 +80,14 @@ func (w *WorkerRelay) OnNewOffer(remoteOfferAnswer *OfferAnswer) {
 	w.relayedConn = relayedConn
 	w.relayLock.Unlock()
 
-	if err := w.relayManager.AddCloseListener(relayedConn.RemoteAddr().String(), w.onRelayClientDisconnected); err != nil {
-		w.log.Errorf("failed to add close listener: %s", err)
+	err = w.relayManager.AddCloseListener(srv, w.onRelayClientDisconnected)
+	if err != nil {
+		log.Errorf("failed to add close listener: %s", err)
 		_ = relayedConn.Close()
 		return
 	}
 
-	w.log.Debugf("peer conn opened via Relay: %s", relayedConn.RemoteAddr())
+	w.log.Debugf("peer conn opened via Relay: %s", srv)
 	go w.conn.onRelayConnectionIsReady(RelayConnInfo{
 		relayedConn:     relayedConn,
 		rosenpassPubKey: remoteOfferAnswer.RosenpassPubKey,
@@ -121,6 +126,13 @@ func (w *WorkerRelay) isRelaySupported(answer *OfferAnswer) bool {
 	return answer.RelaySrvAddress != ""
 }
 
+func (w *WorkerRelay) preferredRelayServer(myRelayAddress, remoteRelayAddress string) string {
+	if w.isController {
+		return myRelayAddress
+	}
+	return remoteRelayAddress
+}
+
 func (w *WorkerRelay) onRelayClientDisconnected() {
 	go w.conn.onRelayDisconnected()
 }

client/internal/profilemanager/config.go

@@ -96,6 +96,7 @@ type ConfigInput struct {
 	BlockLANAccess      *bool
 	BlockInbound        *bool
 	DisableIPv6         *bool
+	SyncMessageVersion  *int
 
 	DisableNotifications *bool
 
@@ -137,6 +138,7 @@ type Config struct {
 	BlockLANAccess      bool
 	BlockInbound        bool
 	DisableIPv6         bool
+	SyncMessageVersion  *int
 
 	DisableNotifications *bool
 
@@ -587,6 +589,12 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
+	if input.SyncMessageVersion != nil && *input.SyncMessageVersion != *config.SyncMessageVersion {
+		log.Infof("setting SyncMessageVersion to %v", *input.SyncMessageVersion)
+		*config.SyncMessageVersion = *input.SyncMessageVersion
+		updated = true
+	}
+
 	if input.DisableNotifications != nil && (config.DisableNotifications == nil || *input.DisableNotifications != *config.DisableNotifications) {
 		if *input.DisableNotifications {
 			log.Infof("disabling notifications")

client/internal/routemanager/dnsinterceptor/handler.go

@@ -22,7 +22,6 @@ import (
 	nbdns "github.com/netbirdio/netbird/client/internal/dns"
 	"github.com/netbirdio/netbird/client/internal/dns/resutil"
 	"github.com/netbirdio/netbird/client/internal/peer"
-	"github.com/netbirdio/netbird/client/internal/peerstore"
 	"github.com/netbirdio/netbird/client/internal/routemanager/common"
 	"github.com/netbirdio/netbird/client/internal/routemanager/fakeip"
 	iface "github.com/netbirdio/netbird/client/internal/routemanager/iface"
@@ -40,6 +39,61 @@ type internalDNATer interface {
 	AddInternalDNATMapping(netip.Addr, netip.Addr) error
 }
 
+// peerAllowedIPs reports the tunnel address a peer is reachable on.
+type peerAllowedIPs interface {
+	AllowedIP(pubKey string) (netip.Addr, bool)
+}
+
+// disposition says where a query for a given record type has to be answered.
+type disposition int
+
+const (
+	// dispositionPeer: the routing peer owns the answer. Address records are
+	// what a DNS route exists for, and their answer programs the routes,
+	// allowed IPs and firewall sets, so they are never resolved anywhere else.
+	dispositionPeer disposition = iota
+	// dispositionPeerFirst: the peer's forwarder resolves the type, and the
+	// records may only exist inside the routed network, so it has to be asked.
+	// A peer running a client that predates that support cannot answer, which
+	// only its reply reveals.
+	dispositionPeerFirst
+	// dispositionChain: no forwarder resolves the type, so asking would cost a
+	// tunnel round trip for a reply that says nothing. Public records for the
+	// name are still better than none, so the query goes to the rest of the
+	// chain.
+	dispositionChain
+)
+
+func dispositionFor(qtype uint16) disposition {
+	switch {
+	case qtype == dns.TypeA || qtype == dns.TypeAAAA:
+		return dispositionPeer
+	case resutil.SupportedRecordQtype(qtype):
+		return dispositionPeerFirst
+	default:
+		return dispositionChain
+	}
+}
+
+// peerCannotAnswer reports whether a forwarder reply means the peer is unable to
+// resolve this query type, as opposed to having resolved it and found nothing.
+// A forwarder older than the one that learned to resolve non-address types
+// answers NOTIMP to all of them, before it even looks at the domain; FORMERR
+// covers one that cannot parse the EDNS0 we add to the query.
+//
+// REFUSED is deliberately not here. It says the peer does not hold the name,
+// which happens while the client and the peer disagree about the route set, and
+// falling through then would hand the name of an internal-only domain to a
+// public resolver. No forwarder version reports a missing record type that way.
+func peerCannotAnswer(rcode int) bool {
+	switch rcode {
+	case dns.RcodeNotImplemented, dns.RcodeFormatError:
+		return true
+	default:
+		return false
+	}
+}
+
 type DnsInterceptor struct {
 	mu                   sync.RWMutex
 	route                *route.Route
@@ -50,7 +104,7 @@ type DnsInterceptor struct {
 	currentPeerKey       string
 	interceptedDomains   domainMap
 	wgInterface          iface.WGIface
-	peerStore            *peerstore.Store
+	peerStore            peerAllowedIPs
 	firewall             firewall.Manager
 	fakeIPManager        *fakeip.Manager
 	forwarderPort        *atomic.Uint32
@@ -226,11 +280,14 @@ func (d *DnsInterceptor) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 		return
 	}
 
-	// All query types for an intercepted domain are forwarded to the peer's
-	// DNS forwarder, which owns the name. Falling through to the system
-	// resolver would let it answer NXDOMAIN for a name it isn't authoritative
-	// for, poisoning the whole name (including the A/AAAA records the route
-	// does serve). The forwarder answers NODATA for types it cannot resolve.
+	qtype := r.Question[0].Qtype
+	dispose := dispositionFor(qtype)
+
+	if dispose == dispositionChain {
+		d.deferToChain(w, r, logger, "unsupported-qtype")
+		return
+	}
+
 	d.mu.RLock()
 	peerKey := d.currentPeerKey
 	d.mu.RUnlock()
@@ -246,35 +303,32 @@ func (d *DnsInterceptor) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 		return
 	}
 
-	if r.Extra == nil {
-		r.MsgHdr.AuthenticatedData = true
-	}
-
-	// Advertise EDNS0 to the forwarder so it may return an Extended DNS Error
-	// describing why a lookup failed. The OPT is stripped from the reply when
-	// the original client did not request EDNS0.
-	hadEdns := r.IsEdns0() != nil
-	if !hadEdns {
-		r.SetEdns0(dns.DefaultMsgSize, false)
-	}
+	query, hadEdns := peerQuery(r)
 
 	upstream := net.JoinHostPort(upstreamIP.String(), strconv.FormatUint(uint64(d.forwarderPort.Load()), 10))
 	ctx, cancel := context.WithTimeout(context.Background(), dnsTimeout)
 	defer cancel()
 
-	reply := d.queryUpstreamDNS(ctx, w, r, upstream, upstreamIP, peerKey, logger)
+	reply := d.queryUpstreamDNS(ctx, w, query, upstream, upstreamIP, peerKey, logger)
 	if reply == nil {
 		return
 	}
 
+	// The peer owns the name but its forwarder cannot resolve this type. No
+	// capability is announced anywhere, so the round trip above is the probe.
+	if dispose == dispositionPeerFirst && peerCannotAnswer(reply.Rcode) {
+		d.deferToChain(w, r, logger, "peer-rcode-"+dns.RcodeToString[reply.Rcode])
+		return
+	}
+
 	if ede, ok := resutil.ExtractEDE(reply); ok {
-		resutil.SetMeta(w, "ede", fmt.Sprintf("%d %s", ede.InfoCode, ede.ExtraText))
+		resutil.SetMeta(w, resutil.MetaKeyEDE, fmt.Sprintf("%d %s", ede.InfoCode, ede.ExtraText))
 	}
 	if !hadEdns {
 		resutil.StripOPT(reply)
 	}
 
-	resutil.SetMeta(w, "peer", peerKey)
+	resutil.SetMeta(w, resutil.MetaKeyPeer, peerKey)
 
 	reply.Id = r.Id
 	if err := d.writeMsg(w, reply, logger); err != nil {
@@ -282,6 +336,30 @@ func (d *DnsInterceptor) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 	}
 }
 
+// deferToChain hands the query to the next handler in the chain and asks the
+// chain to soften the negative verdict of whatever answers instead. The
+// resolvers below cannot prove a name inside the routed network absent, so their
+// NXDOMAIN must not reach the client: it would be cached for the name and every
+// type under it, taking the addresses this route does serve with it.
+func (d *DnsInterceptor) deferToChain(w dns.ResponseWriter, r *dns.Msg, logger *log.Entry, reason string) {
+	logger.Tracef("continuing to next handler for domain=%s type=%s reason=%s",
+		r.Question[0].Name, dns.TypeToString[r.Question[0].Qtype], reason)
+
+	resutil.RequestSoftNegative(w)
+	// Carried to the chain's response log line: without it the answer looks like
+	// it came from the fallthrough resolver on its own.
+	resutil.SetMeta(w, resutil.MetaKeyDeferredBy, "dns-route")
+	resutil.SetMeta(w, resutil.MetaKeyDeferredReason, reason)
+
+	resp := new(dns.Msg)
+	resp.SetRcode(r, dns.RcodeNameError)
+	// Set Zero bit to signal handler chain to continue
+	resp.MsgHdr.Zero = true
+	if err := w.WriteMsg(resp); err != nil {
+		logger.Errorf("failed writing DNS continue response: %v", err)
+	}
+}
+
 func (d *DnsInterceptor) writeDNSError(w dns.ResponseWriter, r *dns.Msg, logger *log.Entry, reason string) {
 	logger.Warnf("failed to query upstream for domain=%s: %s", r.Question[0].Name, reason)
 
@@ -621,3 +699,26 @@ func (d *DnsInterceptor) debugPeerTimeout(peerIP netip.Addr, peerKey string) str
 
 	return fmt.Sprintf(" (peer %s)", nbdns.FormatPeerStatus(&peerState))
 }
+
+// peerQuery builds the query sent to a peer's DNS forwarder and reports whether
+// the client itself advertised EDNS0. EDNS0 is added so the forwarder can return
+// an Extended DNS Error describing an upstream failure, and the OPT is stripped
+// from the reply again when the client did not ask for it. The client's message
+// is left untouched: it may still be handed to the next handler in the chain,
+// and neither the OPT nor the AD bit is ours to put on the wire on its behalf.
+func peerQuery(r *dns.Msg) (query *dns.Msg, hadEdns bool) {
+	query = r.Copy()
+	hadEdns = query.IsEdns0() != nil
+
+	// AD tells the forwarder we understand authenticated data. Only set when the
+	// client sent no additional section of its own, so we never overrule what it
+	// asked for.
+	if len(query.Extra) == 0 {
+		query.MsgHdr.AuthenticatedData = true
+	}
+	if !hadEdns {
+		query.SetEdns0(dns.DefaultMsgSize, false)
+	}
+
+	return query, hadEdns
+}

client/internal/routemanager/dnsinterceptor/handler_test.go

@@ -0,0 +1,399 @@
+package dnsinterceptor
+
+import (
+	"net"
+	"net/netip"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/miekg/dns"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.zx2c4.com/wireguard/tun/netstack"
+
+	"github.com/netbirdio/netbird/client/iface/device"
+	"github.com/netbirdio/netbird/client/iface/wgaddr"
+	"github.com/netbirdio/netbird/client/internal/dns/resutil"
+	"github.com/netbirdio/netbird/client/internal/dns/test"
+	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/route"
+)
+
+// softNegativeWriter records what the handler told the chain: whether to soften
+// a negative verdict from the handlers it defers to, and the metadata that ends
+// up on the chain's response log line.
+type softNegativeWriter struct {
+	test.MockResponseWriter
+	softNegative bool
+	meta         map[resutil.MetaKey]string
+}
+
+func (w *softNegativeWriter) RequestSoftNegative() { w.softNegative = true }
+
+func (w *softNegativeWriter) RequestID() string { return "test" }
+
+func (w *softNegativeWriter) SetMeta(key resutil.MetaKey, value string) {
+	if w.meta == nil {
+		w.meta = make(map[resutil.MetaKey]string)
+	}
+	w.meta[key] = value
+}
+
+// TestServeDNS_QtypeTheForwarderCannotResolve covers the record types no DNS
+// forwarder can answer, because the host resolver exposes no API for them.
+// Asking the peer only burns a tunnel round trip, so the query goes straight to
+// the rest of the chain. No peer key is configured, proving no round trip is
+// attempted.
+func TestServeDNS_QtypeTheForwarderCannotResolve(t *testing.T) {
+	qtypes := []uint16{
+		dns.TypeHTTPS,
+		dns.TypeSVCB,
+		dns.TypeCAA,
+		dns.TypeNAPTR,
+		dns.TypeTLSA,
+		dns.TypeSOA,
+	}
+
+	for _, qtype := range qtypes {
+		t.Run(dns.TypeToString[qtype], func(t *testing.T) {
+			d := &DnsInterceptor{}
+			w := &softNegativeWriter{}
+
+			r := new(dns.Msg)
+			r.SetQuestion("db.example.com.", qtype)
+
+			d.ServeDNS(w, r)
+
+			resp := w.GetLastResponse()
+			require.NotNil(t, resp, "a response must be written")
+			assert.Equal(t, dns.RcodeNameError, resp.Rcode, "chain continuation is signalled as NXDOMAIN")
+			assert.True(t, resp.MsgHdr.Zero, "Zero bit must be set so the chain continues")
+			assert.True(t, w.softNegative,
+				"a downstream NXDOMAIN must be softened, or the fallthrough poisons the routed name")
+			assert.Equal(t, "dns-route", w.meta[resutil.MetaKeyDeferredBy],
+				"the chain's response log line must name us as the handler that stepped aside")
+			assert.Equal(t, "unsupported-qtype", w.meta[resutil.MetaKeyDeferredReason],
+				"the reason must survive to the log line, or the fallthrough is invisible")
+		})
+	}
+}
+
+// TestServeDNS_AddressQtypeNeverFallsThrough locks in the opposite case: A and
+// AAAA are what the route exists for, and the peer is authoritative for them.
+// A missing peer is an error the client must see, never a fallthrough to a
+// resolver that knows nothing about the routed name.
+func TestServeDNS_AddressQtypeNeverFallsThrough(t *testing.T) {
+	for _, qtype := range []uint16{dns.TypeA, dns.TypeAAAA} {
+		t.Run(dns.TypeToString[qtype], func(t *testing.T) {
+			d := &DnsInterceptor{}
+			w := &softNegativeWriter{}
+
+			r := new(dns.Msg)
+			r.SetQuestion("db.example.com.", qtype)
+
+			d.ServeDNS(w, r)
+
+			resp := w.GetLastResponse()
+			require.NotNil(t, resp, "a response must be written")
+			assert.Equal(t, dns.RcodeServerFailure, resp.Rcode, "an unusable route must fail, not fall through")
+			assert.False(t, resp.MsgHdr.Zero, "the chain must not continue for address queries")
+			assert.False(t, w.softNegative, "no fallthrough means nothing to soften")
+		})
+	}
+}
+
+// TestDispositionFor pins the query-type policy, and ties the types we ask a
+// peer about to the types a forwarder can actually resolve, so the two ends
+// cannot drift apart.
+func TestDispositionFor(t *testing.T) {
+	tests := []struct {
+		qtype uint16
+		want  disposition
+	}{
+		{dns.TypeA, dispositionPeer},
+		{dns.TypeAAAA, dispositionPeer},
+		{dns.TypeMX, dispositionPeerFirst},
+		{dns.TypeTXT, dispositionPeerFirst},
+		{dns.TypeNS, dispositionPeerFirst},
+		{dns.TypeSRV, dispositionPeerFirst},
+		{dns.TypeCNAME, dispositionPeerFirst},
+		{dns.TypePTR, dispositionPeerFirst},
+		{dns.TypeHTTPS, dispositionChain},
+		{dns.TypeSVCB, dispositionChain},
+		{dns.TypeCAA, dispositionChain},
+		{dns.TypeNAPTR, dispositionChain},
+		{dns.TypeTLSA, dispositionChain},
+		{dns.TypeDS, dispositionChain},
+		{dns.TypeDNSKEY, dispositionChain},
+		{dns.TypeSOA, dispositionChain},
+		{dns.TypeANY, dispositionChain},
+	}
+
+	for _, tt := range tests {
+		t.Run(dns.TypeToString[tt.qtype], func(t *testing.T) {
+			assert.Equal(t, tt.want, dispositionFor(tt.qtype), "disposition for %s", dns.TypeToString[tt.qtype])
+
+			// Address types are resolved by the forwarder's address path, not by
+			// its record path, so they are deliberately outside
+			// SupportedRecordQtype and their disposition does not depend on it.
+			if tt.want == dispositionPeer {
+				assert.False(t, resutil.SupportedRecordQtype(tt.qtype),
+					"address types take the forwarder's address path")
+				return
+			}
+
+			assert.Equal(t, tt.want == dispositionPeerFirst, resutil.SupportedRecordQtype(tt.qtype),
+				"only types a forwarder resolves may be sent to a peer")
+		})
+	}
+}
+
+// TestPeerCannotAnswer separates a peer that cannot resolve a type from one that
+// resolved it and found nothing. Only the former may be retried elsewhere:
+// treating NODATA or NXDOMAIN as a capability failure would send the client to a
+// public resolver behind the peer's back and prefer a public record over the
+// private one the route exists to reach.
+func TestPeerCannotAnswer(t *testing.T) {
+	cannot := []int{dns.RcodeNotImplemented, dns.RcodeFormatError}
+	for _, rcode := range cannot {
+		assert.True(t, peerCannotAnswer(rcode), "%s means the peer cannot answer", dns.RcodeToString[rcode])
+	}
+
+	// REFUSED belongs here, not above: it says the peer does not hold the name,
+	// so retrying elsewhere would leak the name of an internal-only domain to a
+	// public resolver. No forwarder version reports a missing record type this
+	// way, so nothing is lost by keeping it out.
+	can := []int{
+		dns.RcodeSuccess,
+		dns.RcodeNameError,
+		dns.RcodeServerFailure,
+		dns.RcodeNotAuth,
+		dns.RcodeRefused,
+	}
+	for _, rcode := range can {
+		assert.False(t, peerCannotAnswer(rcode), "%s is the peer's answer, not a capability failure", dns.RcodeToString[rcode])
+	}
+}
+
+// fakeWGIface is the minimum an interceptor needs to reach a peer's DNS
+// forwarder over a plain socket. A nil netstack keeps the exchange on the host
+// stack so the test can point it at a loopback listener.
+type fakeWGIface struct{}
+
+func (fakeWGIface) AddAllowedIP(string, netip.Prefix) error    { return nil }
+func (fakeWGIface) RemoveAllowedIP(string, netip.Prefix) error { return nil }
+func (fakeWGIface) Name() string                               { return "wt0" }
+func (fakeWGIface) Address() wgaddr.Address                    { return wgaddr.Address{} }
+func (fakeWGIface) ToInterface() *net.Interface                { return nil }
+func (fakeWGIface) IsUserspaceBind() bool                      { return false }
+func (fakeWGIface) GetFilter() device.PacketFilter             { return nil }
+func (fakeWGIface) GetDevice() *device.FilteredDevice          { return nil }
+func (fakeWGIface) GetNet() *netstack.Net                      { return nil }
+
+// fakePeerIPs resolves every peer key to the loopback address the test's
+// forwarder stub listens on.
+type fakePeerIPs struct {
+	addr netip.Addr
+}
+
+func (p fakePeerIPs) AllowedIP(string) (netip.Addr, bool) { return p.addr, p.addr.IsValid() }
+
+// forwarderStub stands in for the DNS forwarder of a routing peer, recording
+// what it was asked and answering with a canned reply.
+type forwarderStub struct {
+	mu       sync.Mutex
+	queries  []*dns.Msg
+	reply    func(q *dns.Msg) *dns.Msg
+	port     uint16
+	shutdown func()
+}
+
+func (s *forwarderStub) received() []*dns.Msg {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return append([]*dns.Msg(nil), s.queries...)
+}
+
+func newForwarderStub(t *testing.T, reply func(q *dns.Msg) *dns.Msg) *forwarderStub {
+	t.Helper()
+
+	conn, err := net.ListenUDP("udp", net.UDPAddrFromAddrPort(netip.MustParseAddrPort("127.0.0.1:0")))
+	require.NoError(t, err, "listen for the forwarder stub")
+
+	stub := &forwarderStub{reply: reply}
+	stub.port = uint16(conn.LocalAddr().(*net.UDPAddr).Port)
+
+	mux := dns.NewServeMux()
+	mux.HandleFunc(".", func(w dns.ResponseWriter, q *dns.Msg) {
+		stub.mu.Lock()
+		stub.queries = append(stub.queries, q.Copy())
+		stub.mu.Unlock()
+		_ = w.WriteMsg(stub.reply(q))
+	})
+
+	srv := &dns.Server{PacketConn: conn, Handler: mux}
+	started := make(chan struct{})
+	srv.NotifyStartedFunc = func() { close(started) }
+	go func() {
+		_ = srv.ActivateAndServe()
+	}()
+	select {
+	case <-started:
+	case <-time.After(5 * time.Second):
+		t.Fatal("forwarder stub did not start")
+	}
+
+	stub.shutdown = func() { _ = srv.Shutdown() }
+	t.Cleanup(stub.shutdown)
+
+	return stub
+}
+
+func newTestInterceptor(t *testing.T, stub *forwarderStub) *DnsInterceptor {
+	t.Helper()
+
+	port := new(atomic.Uint32)
+	port.Store(uint32(stub.port))
+
+	return &DnsInterceptor{
+		route:              &route.Route{Domains: nil},
+		statusRecorder:     peer.NewRecorder("https://mgm"),
+		currentPeerKey:     "peer-key",
+		interceptedDomains: make(domainMap),
+		wgInterface:        fakeWGIface{},
+		peerStore:          fakePeerIPs{addr: netip.MustParseAddr("127.0.0.1")},
+		forwarderPort:      port,
+	}
+}
+
+// TestServeDNS_PeerCannotResolveQtype is the reported regression: a routing peer
+// running a client older than the one that taught the forwarder about non-address
+// record types answers NOTIMP for every SRV query, and the interceptor used to
+// hand that straight to the application, breaking SRV-based service discovery.
+// The client cannot know the peer's capability up front, so it must try the peer
+// and fall back to the chain when the peer says it cannot answer.
+func TestServeDNS_PeerCannotResolveQtype(t *testing.T) {
+	for _, rcode := range []int{dns.RcodeNotImplemented, dns.RcodeFormatError} {
+		t.Run(dns.RcodeToString[rcode], func(t *testing.T) {
+			stub := newForwarderStub(t, func(q *dns.Msg) *dns.Msg {
+				resp := new(dns.Msg)
+				resp.SetRcode(q, rcode)
+				return resp
+			})
+			d := newTestInterceptor(t, stub)
+			w := &softNegativeWriter{}
+
+			r := new(dns.Msg)
+			r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+
+			d.ServeDNS(w, r)
+
+			require.Len(t, stub.received(), 1, "the peer must be asked before giving up on it")
+
+			resp := w.GetLastResponse()
+			require.NotNil(t, resp, "a response must be written")
+			assert.Equal(t, dns.RcodeNameError, resp.Rcode, "chain continuation is signalled as NXDOMAIN")
+			assert.True(t, resp.MsgHdr.Zero, "Zero bit must be set so the chain continues")
+			assert.True(t, w.softNegative, "the fallthrough must not be allowed to poison the routed name")
+			assert.Equal(t, "dns-route", w.meta[resutil.MetaKeyDeferredBy])
+			assert.Equal(t, "peer-rcode-"+dns.RcodeToString[rcode], w.meta[resutil.MetaKeyDeferredReason],
+				"the peer's verdict must be visible in the log, it is the only trace of the probe")
+		})
+	}
+}
+
+// TestServeDNS_PeerVerdictPassesThrough covers the replies that are the peer's
+// answer rather than a statement about its capability. The peer is authoritative
+// for a name routed to it, so its verdict reaches the client as it is: retrying
+// these elsewhere would send the name of an internal-only domain to a public
+// resolver and prefer whatever that resolver says over the route.
+func TestServeDNS_PeerVerdictPassesThrough(t *testing.T) {
+	// REFUSED means the peer does not consider the name routed to it, which
+	// happens while the client and the peer disagree about the route set. It is
+	// never how a peer reports a record type it cannot resolve: a forwarder too
+	// old for non-address types answers NOTIMP before it looks at the domain.
+	for _, rcode := range []int{dns.RcodeRefused, dns.RcodeNameError, dns.RcodeSuccess} {
+		t.Run(dns.RcodeToString[rcode], func(t *testing.T) {
+			stub := newForwarderStub(t, func(q *dns.Msg) *dns.Msg {
+				resp := new(dns.Msg)
+				resp.SetRcode(q, rcode)
+				return resp
+			})
+			d := newTestInterceptor(t, stub)
+			w := &softNegativeWriter{}
+
+			r := new(dns.Msg)
+			r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+
+			d.ServeDNS(w, r)
+
+			require.Len(t, stub.received(), 1, "the peer must be asked exactly once")
+
+			resp := w.GetLastResponse()
+			require.NotNil(t, resp, "a response must be written")
+			assert.Equal(t, rcode, resp.Rcode, "the peer's verdict must reach the client unchanged")
+			assert.False(t, resp.MsgHdr.Zero, "the chain must not continue past an answered query")
+			assert.False(t, w.softNegative, "nothing was deferred, so nothing may be softened")
+		})
+	}
+}
+
+// TestServeDNS_PeerResolvesQtype guards the common case: a peer that can answer
+// the type stays authoritative, and its records reach the client unchanged.
+func TestServeDNS_PeerResolvesQtype(t *testing.T) {
+	stub := newForwarderStub(t, func(q *dns.Msg) *dns.Msg {
+		resp := new(dns.Msg)
+		resp.SetReply(q)
+		resp.Answer = []dns.RR{&dns.SRV{
+			Hdr:    dns.RR_Header{Name: q.Question[0].Name, Rrtype: dns.TypeSRV, Class: dns.ClassINET, Ttl: 60},
+			Target: "shard-00.db.example.com.",
+			Port:   27017,
+		}}
+		return resp
+	})
+	d := newTestInterceptor(t, stub)
+	w := &softNegativeWriter{}
+
+	r := new(dns.Msg)
+	r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+
+	d.ServeDNS(w, r)
+
+	resp := w.GetLastResponse()
+	require.NotNil(t, resp, "a response must be written")
+	assert.Equal(t, dns.RcodeSuccess, resp.Rcode)
+	require.Len(t, resp.Answer, 1, "the peer's records must pass through")
+	assert.False(t, resp.MsgHdr.Zero, "an answered query must not continue the chain")
+	assert.False(t, w.softNegative)
+}
+
+// TestServeDNS_DeferredQueryIsPristine covers what the fallthrough hands to the
+// next handler. The interceptor advertises EDNS0 to the peer so the forwarder can
+// return an Extended DNS Error, and sets AD, but neither belongs to the client's
+// query: on the deferred path they would travel to the public resolver and the
+// reply could come back carrying an OPT the client never advertised, which
+// RFC 6891 forbids us from passing on.
+func TestServeDNS_DeferredQueryIsPristine(t *testing.T) {
+	stub := newForwarderStub(t, func(q *dns.Msg) *dns.Msg {
+		resp := new(dns.Msg)
+		resp.SetRcode(q, dns.RcodeNotImplemented)
+		return resp
+	})
+	d := newTestInterceptor(t, stub)
+	w := &softNegativeWriter{}
+
+	r := new(dns.Msg)
+	r.SetQuestion("_mongodb._tcp.db.example.com.", dns.TypeSRV)
+
+	d.ServeDNS(w, r)
+
+	sent := stub.received()
+	require.Len(t, sent, 1)
+	assert.NotNil(t, sent[0].IsEdns0(), "the peer must be asked with EDNS0 so it can return an EDE")
+
+	assert.Nil(t, r.IsEdns0(), "the deferred query must not carry an OPT the client never sent")
+	assert.False(t, r.AuthenticatedData, "the deferred query must not carry an AD bit the client never set")
+	assert.Empty(t, r.Extra, "the deferred query must reach the next handler as the client sent it")
+}

client/internal/routemanager/dnsinterceptor/handler_windows_test.go

@@ -0,0 +1,7 @@
+//go:build windows
+
+package dnsinterceptor
+
+// GetInterfaceGUIDString completes iface.WGIface on Windows, which requires the
+// interface GUID for DNS registration.
+func (fakeWGIface) GetInterfaceGUIDString() (string, error) { return "", nil }

client/internal/routemanager/dynamic/route.go

@@ -185,7 +185,7 @@ func (r *Route) startResolver(ctx context.Context) {
 }
 
 func (r *Route) update(ctx context.Context) error {
-	resolved, err := r.resolveDomains()
+	resolved, err := r.resolveDomains(ctx)
 	if err != nil {
 		if len(resolved) == 0 {
 			return fmt.Errorf("resolve domains: %w", err)
@@ -199,9 +199,9 @@ func (r *Route) update(ctx context.Context) error {
 	return nil
 }
 
-func (r *Route) resolveDomains() (domainMap, error) {
+func (r *Route) resolveDomains(ctx context.Context) (domainMap, error) {
 	results := make(chan resolveResult)
-	go r.resolve(results)
+	go r.resolve(ctx, results)
 
 	resolved := domainMap{}
 	var merr *multierror.Error
@@ -217,7 +217,7 @@ func (r *Route) resolveDomains() (domainMap, error) {
 	return resolved, nberrors.FormatErrorOrNil(merr)
 }
 
-func (r *Route) resolve(results chan resolveResult) {
+func (r *Route) resolve(ctx context.Context, results chan resolveResult) {
 	var wg sync.WaitGroup
 
 	for _, d := range r.route.Domains {
@@ -225,10 +225,10 @@ func (r *Route) resolve(results chan resolveResult) {
 		go func(domain domain.Domain) {
 			defer wg.Done()
 
-			ips, err := r.getIPsFromResolver(domain)
+			ips, err := r.getIPsFromResolver(ctx, domain)
 			if err != nil {
 				log.Tracef("Failed to resolve domain %s with private resolver: %v", domain.SafeString(), err)
-				ips, err = net.LookupIP(domain.PunycodeString())
+				ips, err = lookupHostIPs(ctx, domain)
 				if err != nil {
 					results <- resolveResult{domain: domain, err: fmt.Errorf("resolve d %s: %w", domain.SafeString(), err)}
 					return
@@ -364,6 +364,20 @@ func determinePrefixChanges(oldPrefixes, newPrefixes []netip.Prefix) (toAdd, toR
 	return
 }
 
+// lookupHostIPs resolves d via the system resolver, honoring ctx cancellation.
+func lookupHostIPs(ctx context.Context, d domain.Domain) ([]net.IP, error) {
+	addrs, err := net.DefaultResolver.LookupIPAddr(ctx, d.PunycodeString())
+	if err != nil {
+		return nil, err
+	}
+
+	ips := make([]net.IP, 0, len(addrs))
+	for _, addr := range addrs {
+		ips = append(ips, addr.IP)
+	}
+	return ips, nil
+}
+
 func combinePrefixes(oldPrefixes, removedPrefixes, addedPrefixes []netip.Prefix) []netip.Prefix {
 	prefixSet := make(map[netip.Prefix]struct{})
 	for _, prefix := range oldPrefixes {

client/internal/routemanager/dynamic/route_generic.go

@@ -3,11 +3,12 @@
 package dynamic
 
 import (
+	"context"
 	"net"
 
 	"github.com/netbirdio/netbird/shared/management/domain"
 )
 
-func (r *Route) getIPsFromResolver(domain domain.Domain) ([]net.IP, error) {
-	return net.LookupIP(domain.PunycodeString())
+func (r *Route) getIPsFromResolver(ctx context.Context, domain domain.Domain) ([]net.IP, error) {
+	return lookupHostIPs(ctx, domain)
 }

client/internal/routemanager/dynamic/route_ios.go

@@ -3,6 +3,7 @@
 package dynamic
 
 import (
+	"context"
 	"fmt"
 	"net"
 	"time"
@@ -16,7 +17,7 @@ import (
 
 const dialTimeout = 10 * time.Second
 
-func (r *Route) getIPsFromResolver(domain domain.Domain) ([]net.IP, error) {
+func (r *Route) getIPsFromResolver(ctx context.Context, domain domain.Domain) ([]net.IP, error) {
 	privateClient, err := nbdns.GetClientPrivate(r.wgInterface, r.resolverAddr.Addr(), dialTimeout)
 	if err != nil {
 		return nil, fmt.Errorf("error while creating private client: %s", err)
@@ -32,7 +33,7 @@ func (r *Route) getIPsFromResolver(domain domain.Domain) ([]net.IP, error) {
 		msg := new(dns.Msg)
 		msg.SetQuestion(fqdn, qtype)
 
-		response, _, err := nbdns.ExchangeWithFallback(nil, privateClient, msg, r.resolverAddr.String())
+		response, _, err := nbdns.ExchangeWithFallback(ctx, privateClient, msg, r.resolverAddr.String())
 		if err != nil {
 			if queryErr == nil {
 				queryErr = fmt.Errorf("DNS query for %s (type %d) after %s: %w", domain.SafeString(), qtype, time.Since(startTime), err)

client/internal/routemanager/manager.go

@@ -61,6 +61,7 @@ type Manager interface {
 	InitialRouteRange() []string
 	SetFirewall(firewall.Manager) error
 	SetDNSForwarderPort(port uint16)
+	ReconcilePeerAllowedIPs(peerKey string) error
 	Stop(stateManager *statemanager.Manager)
 }
 
@@ -232,6 +233,30 @@ func (m *DefaultManager) setupRefCounters(useNoop bool) {
 	)
 }
 
+// ReconcilePeerAllowedIPs re-applies every routed allowed IP currently tracked for the peer
+// onto the WireGuard device. The allowed-IP refcounter only calls its AddFunc (which pushes to
+// the device) on a prefix's 0->1 transition, so a peer whose device entry was rebuilt without a
+// matching refcounter change — e.g. a lazy connection cycling through idle->wake, which recreates
+// the WireGuard peer with the overlay /32 only — ends up missing routed prefixes the refcounter
+// still considers installed, and nothing retries. Calling this when the peer's WireGuard entry is
+// (re)created restores convergence. It is add-only and idempotent: AddAllowedIP is update-only, so
+// prefixes are re-added to an existing peer and an absent peer is left untouched.
+func (m *DefaultManager) ReconcilePeerAllowedIPs(peerKey string) error {
+	if m.allowedIPsRefCounter == nil {
+		return nil
+	}
+
+	return m.allowedIPsRefCounter.ReapplyMatching(
+		func(out string) bool { return out == peerKey },
+		func(prefix netip.Prefix) error {
+			if err := m.wgInterface.AddAllowedIP(peerKey, prefix); err != nil {
+				return fmt.Errorf("add allowed IP %s for peer %s: %w", prefix, peerKey, err)
+			}
+			return nil
+		},
+	)
+}
+
 // Init sets up the routing
 func (m *DefaultManager) Init() error {
 	m.routeSelector = m.initSelector()

client/internal/routemanager/mock.go

@@ -112,6 +112,11 @@ func (m *MockManager) SetFirewall(firewall.Manager) error {
 func (m *MockManager) SetDNSForwarderPort(port uint16) {
 }
 
+// ReconcilePeerAllowedIPs mock implementation of ReconcilePeerAllowedIPs from Manager interface
+func (m *MockManager) ReconcilePeerAllowedIPs(peerKey string) error {
+	return nil
+}
+
 // Stop mock implementation of Stop from Manager interface
 func (m *MockManager) Stop(stateManager *statemanager.Manager) {
 	if m.StopFunc != nil {

client/internal/routemanager/reconcile_test.go

@@ -0,0 +1,90 @@
+//go:build !windows
+
+package routemanager
+
+import (
+	"net"
+	"net/netip"
+	"sync"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.zx2c4.com/wireguard/tun/netstack"
+
+	"github.com/netbirdio/netbird/client/iface/device"
+	"github.com/netbirdio/netbird/client/iface/wgaddr"
+	"github.com/netbirdio/netbird/client/internal/routemanager/refcounter"
+)
+
+// reconcileWGMock is a minimal iface.WGIface that only records AddAllowedIP calls; every other
+// method is an inert stub because ReconcilePeerAllowedIPs exercises none of them.
+type reconcileWGMock struct {
+	mu   sync.Mutex
+	adds map[string][]netip.Prefix
+}
+
+func (m *reconcileWGMock) AddAllowedIP(peerKey string, allowedIP netip.Prefix) error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if m.adds == nil {
+		m.adds = map[string][]netip.Prefix{}
+	}
+	m.adds[peerKey] = append(m.adds[peerKey], allowedIP)
+	return nil
+}
+
+func (m *reconcileWGMock) added(peerKey string) []netip.Prefix {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	return m.adds[peerKey]
+}
+
+func (m *reconcileWGMock) RemoveAllowedIP(string, netip.Prefix) error { return nil }
+func (m *reconcileWGMock) Name() string                               { return "utun-test" }
+func (m *reconcileWGMock) Address() wgaddr.Address                    { return wgaddr.Address{} }
+func (m *reconcileWGMock) ToInterface() *net.Interface                { return nil }
+func (m *reconcileWGMock) IsUserspaceBind() bool                      { return false }
+func (m *reconcileWGMock) GetFilter() device.PacketFilter             { return nil }
+func (m *reconcileWGMock) GetDevice() *device.FilteredDevice          { return nil }
+func (m *reconcileWGMock) GetNet() *netstack.Net                      { return nil }
+
+// TestReconcilePeerAllowedIPs verifies the declarative reconcile re-applies every routed prefix
+// tracked for the peer (self-heal, independent of refcount level) and stays scoped to that peer.
+func TestReconcilePeerAllowedIPs(t *testing.T) {
+	wg := &reconcileWGMock{}
+	m := &DefaultManager{wgInterface: wg}
+	m.allowedIPsRefCounter = refcounter.New[netip.Prefix, string, string](
+		func(_ netip.Prefix, peerKey string) (string, error) { return peerKey, nil },
+		func(netip.Prefix, string) error { return nil },
+	)
+
+	peerA1 := netip.MustParsePrefix("10.0.0.0/24")
+	peerA2 := netip.MustParsePrefix("10.1.0.0/24")
+	peerB1 := netip.MustParsePrefix("10.2.0.0/24")
+
+	for prefix, peer := range map[netip.Prefix]string{peerA1: "peerA", peerA2: "peerA", peerB1: "peerB"} {
+		_, err := m.allowedIPsRefCounter.Increment(prefix, peer)
+		require.NoError(t, err)
+	}
+	// Extra reference: reconcile must still re-apply the prefix even though its refcount never
+	// hit 0 again (the exact case the plain incremental path skips).
+	_, err := m.allowedIPsRefCounter.Increment(peerA1, "peerA")
+	require.NoError(t, err)
+
+	require.NoError(t, m.ReconcilePeerAllowedIPs("peerA"))
+
+	assert.ElementsMatch(t, []netip.Prefix{peerA1, peerA2}, wg.added("peerA"),
+		"reconcile must re-apply all routed prefixes of the peer")
+	assert.Empty(t, wg.added("peerB"), "reconcile must not touch another peer's prefixes")
+}
+
+// TestReconcilePeerAllowedIPsNoCounter verifies reconcile is a safe no-op before the refcounter is
+// set up.
+func TestReconcilePeerAllowedIPsNoCounter(t *testing.T) {
+	wg := &reconcileWGMock{}
+	m := &DefaultManager{wgInterface: wg}
+
+	require.NoError(t, m.ReconcilePeerAllowedIPs("peerA"))
+	assert.Empty(t, wg.added("peerA"))
+}

client/internal/routemanager/refcounter/refcounter.go

@@ -94,6 +94,26 @@ func (rm *Counter[Key, I, O]) Get(key Key) (Ref[O], bool) {
 	return ref, ok
 }
 
+// ReapplyMatching calls apply for every key whose stored Out satisfies pred, holding the
+// counter lock for the whole pass. Running apply under the lock keeps it atomic with respect
+// to Increment/Decrement: a prefix dropped to zero is removed from the map (and had its
+// RemoveFunc called) before this pass observes it, so a stale key can never be re-applied.
+// pred and apply are invoked under the lock, so they must not call back into the counter.
+func (rm *Counter[Key, I, O]) ReapplyMatching(pred func(out O) bool, apply func(key Key) error) error {
+	rm.mu.Lock()
+	defer rm.mu.Unlock()
+
+	var merr *multierror.Error
+	for key, ref := range rm.refCountMap {
+		if pred(ref.Out) {
+			if err := apply(key); err != nil {
+				merr = multierror.Append(merr, err)
+			}
+		}
+	}
+	return nberrors.FormatErrorOrNil(merr)
+}
+
 // Increment increments the reference count for the given key.
 // If this is the first reference to the key, the AddFunc is called.
 func (rm *Counter[Key, I, O]) Increment(key Key, in I) (Ref[O], error) {

client/internal/routemanager/refcounter/refcounter_test.go

@@ -0,0 +1,47 @@
+package refcounter
+
+import (
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestReapplyMatching verifies ReapplyMatching invokes apply for exactly the keys whose stored
+// Out satisfies the predicate (no duplicates for multiply-referenced keys) — the primitive
+// ReconcilePeerAllowedIPs relies on to re-apply a single peer's routed prefixes.
+func TestReapplyMatching(t *testing.T) {
+	rc := New[netip.Prefix, string, string](
+		func(_ netip.Prefix, peerKey string) (string, error) { return peerKey, nil },
+		func(netip.Prefix, string) error { return nil },
+	)
+
+	peerA1 := netip.MustParsePrefix("10.0.0.0/24")
+	peerA2 := netip.MustParsePrefix("10.1.0.0/24")
+	peerB1 := netip.MustParsePrefix("10.2.0.0/24")
+
+	for prefix, peer := range map[netip.Prefix]string{peerA1: "peerA", peerA2: "peerA", peerB1: "peerB"} {
+		_, err := rc.Increment(prefix, peer)
+		require.NoError(t, err)
+	}
+	// a second reference must not make the key applied twice
+	_, err := rc.Increment(peerA1, "peerA")
+	require.NoError(t, err)
+
+	var applied []netip.Prefix
+	err = rc.ReapplyMatching(
+		func(out string) bool { return out == "peerA" },
+		func(key netip.Prefix) error { applied = append(applied, key); return nil },
+	)
+	require.NoError(t, err)
+	assert.ElementsMatch(t, []netip.Prefix{peerA1, peerA2}, applied)
+
+	var none []netip.Prefix
+	err = rc.ReapplyMatching(
+		func(out string) bool { return out == "missing" },
+		func(key netip.Prefix) error { none = append(none, key); return nil },
+	)
+	require.NoError(t, err)
+	assert.Empty(t, none)
+}

client/server/server.go

@@ -1081,7 +1081,10 @@ func (s *Server) Down(ctx context.Context, _ *proto.DownRequest) (*proto.DownRes
 
 	if err := s.cleanupConnection(); err != nil {
 		s.mutex.Unlock()
-		// todo review to update the status in case any type of error
+		if errors.Is(err, ErrServiceNotUp) {
+			log.Debugf("Down called while service not up: %v", err)
+			return nil, err
+		}
 		log.Errorf("failed to shut down properly: %v", err)
 		return nil, err
 	}
@@ -1154,7 +1157,7 @@ func (s *Server) cleanupConnection() error {
 	// making the run loop the sole owner of engine shutdown.
 	if engine != nil {
 		if err := engine.Stop(); err != nil {
-			return err
+			log.Errorf("failed to stop engine during cleanup: %v", err)
 		}
 	}
 

client/system/info.go

@@ -79,13 +79,15 @@ type Info struct {
 	EnableSSHLocalPortForwarding  bool
 	EnableSSHRemotePortForwarding bool
 	DisableSSHAuth                bool
+
+	SyncMessageVersion *int
 }
 
 func (i *Info) SetFlags(
 	rosenpassEnabled, rosenpassPermissive bool,
 	serverSSHAllowed *bool,
 	disableClientRoutes, disableServerRoutes,
-	disableDNS, disableFirewall, blockLANAccess, blockInbound, disableIPv6 bool,
+	disableDNS, disableFirewall, blockLANAccess, blockInbound, disableIPv6 bool, syncMessageVersion *int,
 	enableSSHRoot, enableSSHSFTP, enableSSHLocalPortForwarding, enableSSHRemotePortForwarding *bool,
 	disableSSHAuth *bool,
 ) {
@@ -103,6 +105,8 @@ func (i *Info) SetFlags(
 	i.BlockInbound = blockInbound
 	i.DisableIPv6 = disableIPv6
 
+	i.SyncMessageVersion = syncMessageVersion
+
 	if enableSSHRoot != nil {
 		i.EnableSSHRoot = *enableSSHRoot
 	}

client/ui/autostart_default.go

@@ -51,7 +51,7 @@ func autostartDisabledByMDM(policy *mdm.Policy) bool {
 // netbirdFootprintExists reports whether the machine already carries NetBird
 // daemon config or state, meaning this is not a genuinely fresh install. It is
 // the update-safety gate for the autostart default: upgrading users always
-// have a footprint, so an update can never trigger a login-item write.
+// have a footprint, so an update can never trigger a autostart entry write.
 func netbirdFootprintExists() bool {
 	candidates := []string{
 		profilemanager.DefaultConfigPath,
@@ -69,9 +69,23 @@ func netbirdFootprintExists() bool {
 // applyAutostartDefault runs the one-time launch-on-login default for genuinely
 // fresh installs. The autostartInitialized marker is persisted before any
 // enable attempt so a crash mid-flow degrades to "never enabled" instead of
-// retrying login-item writes on every launch. A user's later disable in
+// retrying autostart entry writes on every launch. A user's later disable in
 // Settings is never overridden: the marker guarantees at-most-once, ever.
 func applyAutostartDefault(ctx context.Context, autostart *services.Autostart, prefs *preferences.Store, prefsFileExisted bool) {
+	mdmDisabled := autostartDisabledByMDM(mdm.LoadPolicy())
+
+	if mdmDisabled {
+		if enabled, err := autostart.IsEnabled(ctx); err != nil {
+			log.Warnf("MDM disableAutostart: read autostart state: %v", err)
+		} else if enabled {
+			if err := autostart.SetEnabled(ctx, false); err != nil {
+				log.Warnf("MDM disableAutostart: force off failed: %v", err)
+			} else {
+				log.Info("MDM disableAutostart enforced: autostart turned off")
+			}
+		}
+	}
+
 	priorFootprint := netbirdFootprintExists() || prefsFileExisted
 
 	if prefs.Get().AutostartInitialized {
@@ -84,7 +98,7 @@ func applyAutostartDefault(ctx context.Context, autostart *services.Autostart, p
 
 	state := autostartDefaultState{
 		supported:    autostart.Supported(ctx),
-		mdmDisabled:  autostartDisabledByMDM(mdm.LoadPolicy()),
+		mdmDisabled:  mdmDisabled,
 		priorInstall: priorFootprint,
 	}
 	enable, reason := shouldEnableAutostartDefault(state)

client/ui/frontend/src/components/empty-state/DaemonOutdatedOverlay.tsx

@@ -1,10 +1,13 @@
+import { useEffect, useState } from "react";
 import { useTranslation } from "react-i18next";
 import { AlertTriangleIcon, DownloadIcon } from "lucide-react";
 import { Browser } from "@wailsio/runtime";
+import { Version } from "@bindings/services";
 import { Button } from "@/components/buttons/Button";
 import { useStatus } from "@/contexts/StatusContext.tsx";
 
 const RELEASES_URL = "https://github.com/netbirdio/netbird/releases/latest";
+const RC_RELEASES_URL = "https://pkgs.netbird.io/releases/rc";
 
 function openUrl(url: string) {
     Browser.OpenURL(url).catch(() => globalThis.open(url, "_blank"));
@@ -12,7 +15,26 @@ function openUrl(url: string) {
 
 export const DaemonOutdatedOverlay = () => {
     const { t } = useTranslation();
-    const { isDaemonOutdated } = useStatus();
+    const { status, isDaemonOutdated } = useStatus();
+
+    const [guiVersion, setGuiVersion] = useState<string>("-");
+    const clientVersion = status?.daemonVersion ?? "—";
+
+    const isRc = /-rc/i.test(guiVersion) || /-rc/i.test(clientVersion);
+    const downloadUrl = isRc ? RC_RELEASES_URL : RELEASES_URL;
+
+    useEffect(() => {
+        if (!isDaemonOutdated) return;
+        let cancelled = false;
+        Version.GUI()
+            .then((v) => {
+                if (!cancelled) setGuiVersion(v);
+            })
+            .catch((err) => console.error("[DaemonOutdatedOverlay] GUI version error", err));
+        return () => {
+            cancelled = true;
+        };
+    }, [isDaemonOutdated]);
 
     if (!isDaemonOutdated) return null;
 
@@ -38,10 +60,37 @@ export const DaemonOutdatedOverlay = () => {
                     <p className={"text-sm text-nb-gray-300"}>{t("daemon.outdated.description")}</p>
                 </div>
 
+                <div className={"flex flex-col items-center gap-0.5 text-center"}>
+                    <p className={"text-sm font-semibold text-nb-gray-100"}>
+                        {clientVersion === "development" ? (
+                            <span>
+                                {t("settings.about.clientName")}{" "}
+                                <span className={"font-mono text-yellow-400"}>
+                                    {t("settings.about.development")}
+                                </span>
+                            </span>
+                        ) : (
+                            t("settings.about.client", { version: clientVersion })
+                        )}
+                    </p>
+                    <p className={"text-sm font-medium text-nb-gray-250"}>
+                        {guiVersion === "development" ? (
+                            <span>
+                                {t("settings.about.guiName")}{" "}
+                                <span className={"font-mono text-yellow-400"}>
+                                    {t("settings.about.development")}
+                                </span>
+                            </span>
+                        ) : (
+                            t("settings.about.gui", { version: guiVersion })
+                        )}
+                    </p>
+                </div>
+
                 <div className={"wails-no-draggable"}>
-                    <Button variant={"primary"} size={"xs"} onClick={() => openUrl(RELEASES_URL)}>
+                    <Button variant={"primary"} size={"xs"} onClick={() => openUrl(downloadUrl)}>
                         <DownloadIcon size={14} />
-                        {t("update.card.getInstaller")}
+                        {t("daemon.outdated.download")}
                     </Button>
                 </div>
             </div>

client/ui/frontend/src/contexts/ProfileContext.tsx

@@ -28,6 +28,7 @@ type ProfileContextValue = {
     loaded: boolean;
     refresh: () => Promise<void>;
     switchProfile: (id: string) => Promise<void>;
+    switchProfileNoConnect: (id: string) => Promise<void>;
     addProfile: (name: string) => Promise<string>;
     removeProfile: (id: string) => Promise<void>;
     renameProfile: (id: string, newName: string) => Promise<void>;
@@ -112,6 +113,16 @@ export const ProfileProvider = ({ children }: { children: ReactNode }) => {
         [username, refresh],
     );
 
+    // Manage-profiles variant: switches without connecting, so the user can
+    // still adjust the management URL before bringing the connection up.
+    const switchProfileNoConnect = useCallback(
+        async (id: string) => {
+            await ProfileSwitcher.SwitchActiveNoConnect({ profileName: id, username });
+            await refresh();
+        },
+        [username, refresh],
+    );
+
     // addProfile creates a profile by display name and returns the
     // daemon-generated ID, so the caller can immediately address it by ID.
     const addProfile = useCallback(
@@ -158,6 +169,7 @@ export const ProfileProvider = ({ children }: { children: ReactNode }) => {
             loaded,
             refresh,
             switchProfile,
+            switchProfileNoConnect,
             addProfile,
             removeProfile,
             renameProfile,
@@ -171,6 +183,7 @@ export const ProfileProvider = ({ children }: { children: ReactNode }) => {
             loaded,
             refresh,
             switchProfile,
+            switchProfileNoConnect,
             addProfile,
             removeProfile,
             renameProfile,

client/ui/frontend/src/modules/profiles/ProfilesTab.tsx

@@ -45,7 +45,7 @@ export function ProfilesTab() {
         activeProfileId,
         loaded,
         username,
-        switchProfile,
+        switchProfileNoConnect,
         addProfile,
         removeProfile,
         renameProfile,
@@ -100,7 +100,7 @@ export function ProfilesTab() {
             confirmLabel: t("profile.switch.confirm"),
         });
         if (!ok) return;
-        await guarded(i18next.t("profile.error.switchTitle"), () => switchProfile(id));
+        await guarded(i18next.t("profile.error.switchTitle"), () => switchProfileNoConnect(id));
     };
 
     const handleDeregister = async (id: string, name: string) => {
@@ -129,14 +129,13 @@ export function ProfilesTab() {
         await guarded(i18next.t("profile.error.createTitle"), async () => {
             const id = await addProfile(name);
             // SetConfig is keyed by the new profile's ID, so it writes the
-            // not-yet-active profile. Write before switching so any reconnect
-            // targets the right deployment.
+            // not-yet-active profile before the switch makes it current.
             if (!isNetbirdCloud(managementUrl)) {
                 await SettingsSvc.SetConfig(
                     new SetConfigParams({ profileName: id, username, managementUrl }),
                 );
             }
-            await switchProfile(id);
+            await switchProfileNoConnect(id);
         });
     };
 

client/ui/frontend/src/modules/session/SessionExpirationDialog.tsx

@@ -73,6 +73,13 @@ export default function SessionExpirationDialog() {
 
         let offCancel: (() => void) | undefined;
 
+        // Return the dialog to its interactive state and dismiss the browser popup
+        const resetDialog = () => {
+            offCancel?.();
+            WindowManager.CloseBrowserLogin().catch(console.error);
+            setBusy(false);
+        };
+
         try {
             const start = await Session.RequestExtend({ hint: "" });
             const uri = start.verificationUriComplete || start.verificationUri;
@@ -105,25 +112,22 @@ export default function SessionExpirationDialog() {
             if (outcome.kind === "cancel") {
                 waitPromise.cancel?.();
                 waitPromise.catch(() => {});
+                resetDialog();
                 return;
             }
 
             // Another surface owns this flow; keep the dialog open to retry.
             if (outcome.result.preempted) {
+                resetDialog();
                 return;
             }
-
-            // Close before the popup so the restore can't flash this window back.
-            WindowManager.CloseSessionExpiration().catch(console.error);
+            WindowManager.CloseRenewFlow().catch(console.error);
         } catch (e) {
+            resetDialog();
             await errorDialog({
                 Title: t("sessionExpiration.extendFailedTitle"),
                 Message: formatErrorMessage(e),
             });
-        } finally {
-            offCancel?.();
-            WindowManager.CloseBrowserLogin().catch(console.error);
-            setBusy(false);
         }
     }, [busy, t]);
 
@@ -139,12 +143,11 @@ export default function SessionExpirationDialog() {
             });
             WindowManager.CloseSessionExpiration().catch(console.error);
         } catch (e) {
+            setBusy(false);
             await errorDialog({
                 Title: t("sessionExpiration.logoutFailedTitle"),
                 Message: formatErrorMessage(e),
             });
-        } finally {
-            setBusy(false);
         }
     }, [busy, t]);
 

client/ui/frontend/src/modules/welcome/WelcomeStepTray.tsx

@@ -22,6 +22,9 @@ type WelcomeStepTrayProps = {
 export function WelcomeStepTray({ onContinue }: Readonly<WelcomeStepTrayProps>) {
     const { t } = useTranslation();
     const trayScreenshot = trayScreenshotForOS();
+    // macOS has no tray — the icon sits in the menu bar, so the copy says so.
+    const titleKey = isMacOS() ? "welcome.titleMac" : "welcome.title";
+    const descriptionKey = isMacOS() ? "welcome.descriptionMac" : "welcome.description";
 
     return (
         <>
@@ -36,9 +39,9 @@ export function WelcomeStepTray({ onContinue }: Readonly<WelcomeStepTrayProps>)
 
             <div className={"flex w-full flex-col gap-1"}>
                 <DialogHeading id={"nb-welcome-title"} align={"left"}>
-                    {t("welcome.title")}
+                    {t(titleKey)}
                 </DialogHeading>
-                <DialogDescription align={"left"}>{t("welcome.description")}</DialogDescription>
+                <DialogDescription align={"left"}>{t(descriptionKey)}</DialogDescription>
             </div>
 
             <DialogActions>

client/ui/i18n/locales/de/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Suchen Sie NetBird in der Taskleiste"
     },
+    "welcome.titleMac": {
+        "message": "Suchen Sie NetBird in der Menüleiste"
+    },
     "welcome.description": {
         "message": "NetBird läuft in Ihrer Taskleiste. Klicken Sie auf das Symbol, um sich zu verbinden, Profile zu wechseln oder die Einstellungen zu öffnen."
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird läuft in Ihrer Menüleiste. Klicken Sie auf das Symbol, um sich zu verbinden, Profile zu wechseln oder die Einstellungen zu öffnen."
+    },
     "welcome.continue": {
         "message": "Weiter"
     },
@@ -1293,10 +1299,13 @@
         "message": "Dokumentation"
     },
     "daemon.outdated.title": {
-        "message": "NetBird-Dienst ist veraltet"
+        "message": "NetBird Client ist veraltet"
     },
     "daemon.outdated.description": {
-        "message": "Aktualisieren Sie den NetBird-Dienst, um diese App zu verwenden."
+        "message": "Die neue GUI ist nicht mit Ihrem älteren Client kompatibel. Aktualisieren Sie Ihren Client, um die neue Anwendung zu verwenden."
+    },
+    "daemon.outdated.download": {
+        "message": "Neueste Version herunterladen"
     },
     "error.jwt_clock_skew": {
         "message": "Anmeldung fehlgeschlagen: Die Uhr dieses Geräts ist nicht mit dem Server synchron. Bitte synchronisieren Sie die Systemuhr und versuchen Sie es erneut."

client/ui/i18n/locales/en/common.json

@@ -1377,11 +1377,19 @@
     },
     "welcome.title": {
         "message": "Look for NetBird in your tray",
-        "description": "Heading on the first onboarding step, pointing the user to the tray icon. 'tray' = system tray / menu bar."
+        "description": "Heading on the first onboarding step, pointing the user to the tray icon. Shown on Windows and Linux; macOS uses welcome.titleMac."
+    },
+    "welcome.titleMac": {
+        "message": "Look for NetBird in your menu bar",
+        "description": "Heading on the first onboarding step on macOS, pointing the user to the menu bar icon. Use your language's Apple term for the macOS menu bar."
     },
     "welcome.description": {
         "message": "NetBird lives in your tray. Click the icon to connect, switch profiles, or open settings.",
-        "description": "Body of the first onboarding step explaining the tray icon."
+        "description": "Body of the first onboarding step explaining the tray icon. Shown on Windows and Linux; macOS uses welcome.descriptionMac."
+    },
+    "welcome.descriptionMac": {
+        "message": "NetBird lives in your menu bar. Click the icon to connect, switch profiles, or open settings.",
+        "description": "Body of the first onboarding step on macOS explaining the menu bar icon. Use your language's Apple term for the macOS menu bar."
     },
     "welcome.continue": {
         "message": "Continue",
@@ -1724,12 +1732,16 @@
         "description": "Documentation link on the daemon-unavailable overlay."
     },
     "daemon.outdated.title": {
-        "message": "NetBird Service Is Outdated",
-        "description": "Title of the overlay shown when the NetBird background service is too old to drive this UI."
+        "message": "NetBird Client Is Outdated",
+        "description": "Title of the overlay shown when the NetBird client (daemon) is too old to drive this UI."
     },
     "daemon.outdated.description": {
-        "message": "Update the NetBird service to use this app.",
-        "description": "Body of the daemon-outdated overlay telling the user to upgrade the service."
+        "message": "The new GUI isn't compatible with the older NetBird client. Update your client to use the new application.",
+        "description": "Body of the daemon-outdated overlay explaining that the GUI is newer than the client and the client must be updated."
+    },
+    "daemon.outdated.download": {
+        "message": "Download Latest",
+        "description": "Button on the daemon-outdated overlay that opens the download page for the latest release."
     },
     "error.jwt_clock_skew": {
         "message": "Sign-in failed: this device's clock is out of sync with the server. Please sync your system clock and try again.",

client/ui/i18n/locales/es/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Busque NetBird en su bandeja del sistema"
     },
+    "welcome.titleMac": {
+        "message": "Busque NetBird en su barra de menús"
+    },
     "welcome.description": {
         "message": "NetBird reside en su bandeja del sistema. Haga clic en el icono para conectarse, cambiar de perfil o abrir la configuración."
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird reside en su barra de menús. Haga clic en el icono para conectarse, cambiar de perfil o abrir la configuración."
+    },
     "welcome.continue": {
         "message": "Continuar"
     },
@@ -1293,10 +1299,13 @@
         "message": "Documentación"
     },
     "daemon.outdated.title": {
-        "message": "El servicio de NetBird está desactualizado"
+        "message": "NetBird Client está desactualizado"
     },
     "daemon.outdated.description": {
-        "message": "Actualice el servicio de NetBird para usar esta aplicación."
+        "message": "La nueva GUI no es compatible con su cliente anterior. Actualice su cliente para usar la nueva aplicación."
+    },
+    "daemon.outdated.download": {
+        "message": "Descargar la última versión"
     },
     "error.jwt_clock_skew": {
         "message": "Error al iniciar sesión: el reloj de este dispositivo no está sincronizado con el servidor. Sincronice el reloj del sistema e inténtelo de nuevo."

client/ui/i18n/locales/fr/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Cherchez NetBird dans votre barre d’état système"
     },
+    "welcome.titleMac": {
+        "message": "Cherchez NetBird dans votre barre des menus"
+    },
     "welcome.description": {
         "message": "NetBird se trouve dans votre barre d’état système. Cliquez sur l’icône pour vous connecter, changer de profil ou ouvrir les paramètres."
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird se trouve dans votre barre des menus. Cliquez sur l’icône pour vous connecter, changer de profil ou ouvrir les paramètres."
+    },
     "welcome.continue": {
         "message": "Continuer"
     },
@@ -1293,10 +1299,13 @@
         "message": "Documentation"
     },
     "daemon.outdated.title": {
-        "message": "Le service NetBird est obsolète"
+        "message": "Le Client NetBird est obsolète"
     },
     "daemon.outdated.description": {
-        "message": "Mettez à jour le service NetBird pour utiliser cette application."
+        "message": "La nouvelle GUI n'est pas compatible avec votre ancien client. Mettez à jour votre client pour utiliser la nouvelle application."
+    },
+    "daemon.outdated.download": {
+        "message": "Télécharger la dernière version"
     },
     "error.jwt_clock_skew": {
         "message": "Échec de la connexion : l’horloge de cet appareil n’est pas synchronisée avec le serveur. Veuillez synchroniser l’horloge de votre système et réessayer."

client/ui/i18n/locales/hu/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Keresse a NetBirdöt a tálcán"
     },
+    "welcome.titleMac": {
+        "message": "Keresse a NetBirdöt a menüsorban"
+    },
     "welcome.description": {
         "message": "A NetBird a tálcán fut. Kattintson az ikonra a csatlakozáshoz, profilváltáshoz vagy a beállítások megnyitásához."
     },
+    "welcome.descriptionMac": {
+        "message": "A NetBird a menüsorban fut. Kattintson az ikonra a csatlakozáshoz, profilváltáshoz vagy a beállítások megnyitásához."
+    },
     "welcome.continue": {
         "message": "Folytatás"
     },
@@ -1293,10 +1299,13 @@
         "message": "Dokumentáció"
     },
     "daemon.outdated.title": {
-        "message": "A NetBird szolgáltatás elavult"
+        "message": "A NetBird Kliens elavult"
     },
     "daemon.outdated.description": {
-        "message": "Frissítsd a NetBird szolgáltatást az alkalmazás használatához."
+        "message": "Az új GUI nem kompatibilis a régebbi klienseddel. Frissítsd a klienst az új alkalmazás használatához."
+    },
+    "daemon.outdated.download": {
+        "message": "Legújabb letöltése"
     },
     "error.jwt_clock_skew": {
         "message": "A bejelentkezés sikertelen: az eszköz órája eltér a szerverétől. Kérjük, szinkronizálja a rendszer óráját, majd próbálja újra."

client/ui/i18n/locales/it/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Cerchi NetBird nella tray"
     },
+    "welcome.titleMac": {
+        "message": "Cerchi NetBird nella barra dei menu"
+    },
     "welcome.description": {
         "message": "NetBird risiede nella tray. Clicchi sull'icona per connettersi, cambiare profilo o aprire le impostazioni."
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird risiede nella barra dei menu. Clicchi sull'icona per connettersi, cambiare profilo o aprire le impostazioni."
+    },
     "welcome.continue": {
         "message": "Continua"
     },
@@ -1293,10 +1299,13 @@
         "message": "Documentazione"
     },
     "daemon.outdated.title": {
-        "message": "Il servizio NetBird è obsoleto"
+        "message": "NetBird Client è obsoleto"
     },
     "daemon.outdated.description": {
-        "message": "Aggiorna il servizio NetBird per usare questa app."
+        "message": "La nuova GUI non è compatibile con il tuo client precedente. Aggiorna il client per usare la nuova applicazione."
+    },
+    "daemon.outdated.download": {
+        "message": "Scarica l'ultima versione"
     },
     "error.jwt_clock_skew": {
         "message": "Accesso non riuscito: l'orologio di questo dispositivo non è sincronizzato con il server. Sincronizzi l'orologio di sistema e riprovi."

client/ui/i18n/locales/ja/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "トレイの NetBird を確認してください"
     },
+    "welcome.titleMac": {
+        "message": "メニューバーの NetBird を確認してください"
+    },
     "welcome.description": {
         "message": "NetBird はトレイに常駐します。アイコンをクリックして、接続、プロファイルの切り替え、設定を開くことができます。"
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird はメニューバーに常駐します。アイコンをクリックして、接続、プロファイルの切り替え、設定を開くことができます。"
+    },
     "welcome.continue": {
         "message": "続ける"
     },

client/ui/i18n/locales/pt/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Procure o NetBird na sua bandeja"
     },
+    "welcome.titleMac": {
+        "message": "Procure o NetBird na sua barra de menus"
+    },
     "welcome.description": {
         "message": "O NetBird fica na sua bandeja. Clique no ícone para conectar, alternar perfis ou abrir as configurações."
     },
+    "welcome.descriptionMac": {
+        "message": "O NetBird fica na sua barra de menus. Clique no ícone para conectar, alternar perfis ou abrir as configurações."
+    },
     "welcome.continue": {
         "message": "Continuar"
     },
@@ -1293,10 +1299,13 @@
         "message": "Documentação"
     },
     "daemon.outdated.title": {
-        "message": "O serviço NetBird está desatualizado"
+        "message": "O NetBird Client está desatualizado"
     },
     "daemon.outdated.description": {
-        "message": "Atualize o serviço NetBird para usar este aplicativo."
+        "message": "A nova GUI não é compatível com o seu cliente mais antigo. Atualize o seu cliente para usar o novo aplicativo."
+    },
+    "daemon.outdated.download": {
+        "message": "Baixar a versão mais recente"
     },
     "error.jwt_clock_skew": {
         "message": "Falha no login: o relógio deste dispositivo está fora de sincronia com o servidor. Sincronize o relógio do sistema e tente novamente."

client/ui/i18n/locales/ru/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "Найдите NetBird в системном трее"
     },
+    "welcome.titleMac": {
+        "message": "Найдите NetBird в строке меню"
+    },
     "welcome.description": {
         "message": "NetBird находится в системном трее. Нажмите на значок, чтобы подключиться, переключить профиль или открыть настройки."
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird находится в строке меню. Нажмите на значок, чтобы подключиться, переключить профиль или открыть настройки."
+    },
     "welcome.continue": {
         "message": "Продолжить"
     },
@@ -1293,10 +1299,13 @@
         "message": "Документация"
     },
     "daemon.outdated.title": {
-        "message": "Служба NetBird устарела"
+        "message": "Клиент NetBird устарел"
     },
     "daemon.outdated.description": {
-        "message": "Обновите службу NetBird, чтобы использовать это приложение."
+        "message": "Новый GUI несовместим с вашим более старым клиентом. Обновите клиент, чтобы использовать новое приложение."
+    },
+    "daemon.outdated.download": {
+        "message": "Скачать последнюю версию"
     },
     "error.jwt_clock_skew": {
         "message": "Не удалось войти: часы этого устройства рассинхронизированы с сервером. Синхронизируйте системные часы и повторите попытку."

client/ui/i18n/locales/zh-CN/common.json

@@ -1034,9 +1034,15 @@
     "welcome.title": {
         "message": "在托盘中查找 NetBird"
     },
+    "welcome.titleMac": {
+        "message": "在菜单栏中查找 NetBird"
+    },
     "welcome.description": {
         "message": "NetBird 驻留在您的托盘中。点击图标即可连接、切换配置文件或打开设置。"
     },
+    "welcome.descriptionMac": {
+        "message": "NetBird 驻留在您的菜单栏中。点击图标即可连接、切换配置文件或打开设置。"
+    },
     "welcome.continue": {
         "message": "继续"
     },
@@ -1293,10 +1299,13 @@
         "message": "文档"
     },
     "daemon.outdated.title": {
-        "message": "NetBird 服务版本过旧"
+        "message": "NetBird 客户端版本过旧"
     },
     "daemon.outdated.description": {
-        "message": "请更新 NetBird 服务以使用此应用。"
+        "message": "新版 GUI 与您较旧的客户端不兼容。请更新客户端以使用新应用。"
+    },
+    "daemon.outdated.download": {
+        "message": "下载最新版本"
     },
     "error.jwt_clock_skew": {
         "message": "登录失败：此设备的时钟与服务器不同步。请同步您的系统时钟后重试。"

client/ui/services/autostart.go

@@ -10,7 +10,7 @@ import (
 	"github.com/wailsapp/wails/v3/pkg/application"
 )
 
-// Autostart facade over Wails' AutostartManager. The OS login-item registration
+// Autostart facade over Wails' AutostartManager. The OS autostart entry registration
 // is the single source of truth; nothing is mirrored to preferences.
 type Autostart struct {
 	mgr *application.AutostartManager

client/ui/services/profileswitcher.go

@@ -12,13 +12,15 @@ import (
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
 )
 
-// ProfileSwitcher holds the reconnect policy shared by the tray and React
-// frontend so both flip profiles identically. The policy keys off prevStatus
-// from DaemonFeed.Get at SwitchActive entry:
+// ProfileSwitcher holds the switch policy shared by the tray and React
+// frontend so both flip profiles identically. SwitchActive (plain selection:
+// header dropdown, tray submenu) always connects after the switch;
+// SwitchActiveNoConnect (manage-profiles screen) never does, so the user can
+// still adjust the management URL before connecting. prevStatus from
+// DaemonFeed.Get at entry only decides the teardown:
 //
-//	Connected/Connecting → Switch + Down + Up; optimistic Connecting paint.
-//	NeedsLogin/LoginFailed/SessionExpired → Switch + Down; clear stale error for re-login.
-//	Idle → Switch only.
+//	Connected/Connecting/NeedsLogin/LoginFailed/SessionExpired → Down first.
+//	Idle → no Down.
 type ProfileSwitcher struct {
 	profiles   *Profiles
 	connection *Connection
@@ -29,29 +31,40 @@ func NewProfileSwitcher(profiles *Profiles, connection *Connection, feed *Daemon
 	return &ProfileSwitcher{profiles: profiles, connection: connection, feed: feed}
 }
 
-// SwitchActive switches to the named profile applying the reconnect policy.
+// SwitchActive switches to the named profile and always connects afterwards.
 func (s *ProfileSwitcher) SwitchActive(ctx context.Context, p ProfileRef) error {
+	return s.switchActive(ctx, p, true)
+}
+
+// SwitchActiveNoConnect switches to the named profile without connecting,
+// tearing down any existing connection first.
+func (s *ProfileSwitcher) SwitchActiveNoConnect(ctx context.Context, p ProfileRef) error {
+	return s.switchActive(ctx, p, false)
+}
+
+func (s *ProfileSwitcher) switchActive(ctx context.Context, p ProfileRef, connect bool) error {
 	prevStatus := ""
-	if st, err := s.feed.Get(ctx); err == nil {
-		prevStatus = st.Status
-	} else {
-		log.Warnf("profileswitcher: get status: %v", err)
+	if s.feed != nil {
+		if st, err := s.feed.Get(ctx); err == nil {
+			prevStatus = st.Status
+		} else {
+			log.Warnf("profileswitcher: get status: %v", err)
+		}
 	}
 
-	wasActive := strings.EqualFold(prevStatus, StatusConnected) ||
-		strings.EqualFold(prevStatus, StatusConnecting)
-	needsDown := wasActive ||
+	needsDown := strings.EqualFold(prevStatus, StatusConnected) ||
+		strings.EqualFold(prevStatus, StatusConnecting) ||
 		strings.EqualFold(prevStatus, StatusNeedsLogin) ||
 		strings.EqualFold(prevStatus, StatusLoginFailed) ||
 		strings.EqualFold(prevStatus, StatusSessionExpired)
 
-	log.Infof("profileswitcher: switch profile=%q prevStatus=%q wasActive=%v needsDown=%v",
-		p.ProfileName, prevStatus, wasActive, needsDown)
+	log.Infof("profileswitcher: switch profile=%q prevStatus=%q connect=%v needsDown=%v",
+		p.ProfileName, prevStatus, connect, needsDown)
 
-	// Optimistic Connecting paint only when wasActive: those prevStatuses emit
-	// stale Connected + transient Idle pushes during Down that must be
-	// suppressed until Up resumes the stream (see DaemonFeed suppression table).
-	if wasActive {
+	// Optimistic Connecting paint plus stale-push suppression during Down (see
+	// DaemonFeed suppression table); also arms the login-watch that pops
+	// browser-login when the new profile turns out to need SSO.
+	if connect && s.feed != nil {
 		s.feed.BeginProfileSwitch()
 	}
 
@@ -76,9 +89,9 @@ func (s *ProfileSwitcher) SwitchActive(ctx context.Context, p ProfileRef) error
 		}
 	}
 
-	if wasActive {
+	if connect {
 		if err := s.connection.Up(ctx, UpParams(p)); err != nil {
-			return fmt.Errorf("reconnect %q: %w", p.ProfileName, err)
+			return fmt.Errorf("connect %q: %w", p.ProfileName, err)
 		}
 	}
 

client/ui/services/windowmanager.go

@@ -185,37 +185,38 @@ func (s *WindowManager) OpenBrowserLogin(uri string) {
 			startURL = "/#/dialog/browser-login?uri=" + url.QueryEscape(uri)
 		}
 		s.hideOtherWindowsLocked("browser-login")
-		// Prefer the main window's screen (multi-monitor); falls back to OS-default centering.
-		var screen *application.Screen
-		if s.mainWindow != nil {
-			if sc, err := s.mainWindow.GetScreen(); err == nil {
-				screen = sc
-			}
-		}
 		opts := DialogWindowOptions("browser-login", s.title("window.title.signIn"), startURL, s.linuxIcon)
 		// Not always-on-top: it would obscure the browser tab the user logs in through.
 		opts.AlwaysOnTop = false
 		opts.InitialPosition = application.WindowCentered
-		opts.Screen = screen
+		// Open on the active (where users cursor is) display, like the session-expiration dialog.
+		opts.Screen = s.getScreenBasedOnCursorPosition()
 		s.browserLogin = s.app.Window.NewWithOptions(opts)
 		bl := s.browserLogin
-		// Red-X close means cancel: emit the event so startLogin() tears down the SSO wait.
 		bl.OnWindowEvent(events.Common.WindowClosing, func(_ *application.WindowEvent) {
-			s.app.Event.Emit(EventBrowserLoginCancel)
 			s.mu.Lock()
-			s.browserLogin = nil
-			s.restoreHiddenWindowsLocked()
+			// Only a live user red-X still has this registered; programmatic closers
+			// nil s.browserLogin first and clean up themselves. Guarding here stops a
+			// stale close event from wiping a replacement popup's state.
+			userClosed := s.browserLogin == bl
+			if userClosed {
+				s.browserLogin = nil
+				s.restoreHiddenWindowsLocked()
+			}
 			s.mu.Unlock()
+			if userClosed {
+				s.app.Event.Emit(EventBrowserLoginCancel)
+			}
 		})
-		s.centerWhenReady(s.browserLogin)
+		s.centerOnCursorScreen(s.browserLogin)
 		return
 	}
 	if uri != "" {
 		s.browserLogin.SetURL("/#/dialog/browser-login?uri=" + url.QueryEscape(uri))
 	}
+	s.centerOnCursorScreen(s.browserLogin)
 	s.browserLogin.Show()
 	s.browserLogin.Focus()
-	s.centerWhenReady(s.browserLogin)
 }
 
 // BrowserLoginWindow returns the live SSO popup, or nil. While non-nil it is the
@@ -238,6 +239,15 @@ func (s *WindowManager) CloseBrowserLogin() {
 	s.mu.Lock()
 	w := s.browserLogin
 	s.browserLogin = nil
+	// The WindowClosing hook no-ops on a programmatic close, so restore here —
+	// but only if a popup was actually open. The frontend calls this even when no
+	// popup was ever shown (e.g. resetDialog() after an early RequestExtend failure,
+	// or connection.ts's catch path), and hiddenForLogin is shared with
+	// OpenInstallProgress, so an unconditional restore could re-show windows a
+	// still-running install-progress is hiding.
+	if w != nil {
+		s.restoreHiddenWindowsLocked()
+	}
 	s.mu.Unlock()
 	if w != nil {
 		w.Close()
@@ -279,6 +289,35 @@ func (s *WindowManager) CloseSessionExpiration() {
 	}
 }
 
+// CloseRenewFlow tears down the SSO session-renewal UI in a single call: it
+// closes the browser-login popup and the session-expiration window together.
+func (s *WindowManager) CloseRenewFlow() {
+	s.mu.Lock()
+	bl := s.browserLogin
+	se := s.sessionExpiration
+	s.browserLogin = nil
+	s.sessionExpiration = nil
+	if se != nil {
+		kept := s.hiddenForLogin[:0]
+		for _, w := range s.hiddenForLogin {
+			if w != se {
+				kept = append(kept, w)
+			}
+		}
+		s.hiddenForLogin = kept
+	}
+	s.restoreHiddenWindowsLocked()
+	s.mu.Unlock()
+
+	// Close after unlock so the re-entrant handlers can take s.mu.
+	if bl != nil {
+		bl.Close()
+	}
+	if se != nil {
+		se.Close()
+	}
+}
+
 // OpenInstallProgress shows the install-progress window and hides the rest for the duration
 // (restored on close). It owns its own result polling since the daemon restarts mid-install.
 func (s *WindowManager) OpenInstallProgress(version string) {

client/ui/tray.go

@@ -30,6 +30,8 @@ const (
 
 	statusError = "Error"
 
+	quitDownTimeout = 5 * time.Second
+
 	urlGitHubRepo     = "https://github.com/netbirdio/netbird"
 	urlGitHubReleases = "https://github.com/netbirdio/netbird/releases/latest"
 	urlDocs           = "https://docs.netbird.io"
@@ -315,8 +317,7 @@ func (t *Tray) relayoutMenu() {
 		if sessionDeadline.IsZero() {
 			t.sessionExpiresItem.SetHidden(true)
 		} else {
-			remaining := t.formatSessionRemaining(time.Until(sessionDeadline))
-			t.sessionExpiresItem.SetLabel(t.loc.T("tray.session.expiresIn", "remaining", remaining))
+			t.sessionExpiresItem.SetLabel(t.sessionRowLabel(sessionDeadline))
 			t.sessionExpiresItem.SetHidden(false)
 		}
 	}
@@ -446,11 +447,28 @@ func (t *Tray) buildMenu() *application.Menu {
 	menu.AddSeparator()
 	menu.Add(t.loc.T("tray.menu.quit")).
 		SetAccelerator("CmdOrCtrl+Q").
-		OnClick(func(*application.Context) { t.app.Quit() })
+		OnClick(func(*application.Context) { t.handleQuit() })
 
 	return menu
 }
 
+func (t *Tray) handleQuit() {
+	t.profileMu.Lock()
+	if t.switchCancel != nil {
+		t.switchCancel()
+		t.switchCancel = nil
+	}
+	t.profileMu.Unlock()
+	t.svc.DaemonFeed.CancelProfileSwitch()
+
+	ctx, cancel := context.WithTimeout(context.Background(), quitDownTimeout)
+	defer cancel()
+	if err := t.svc.Connection.Down(ctx); err != nil {
+		log.Errorf("disconnect on quit: %v", err)
+	}
+	t.app.Quit()
+}
+
 // handleConnect receives the clicked item from the buildMenu closure —
 // t.upItem is menuMu-guarded and must not be read here.
 func (t *Tray) handleConnect(upItem *application.MenuItem) {

client/ui/tray_session.go

@@ -64,11 +64,42 @@ func (t *Tray) applySessionExpiry(deadline *time.Time, connected bool) bool {
 	return changed
 }
 
-// runSessionExpiryTicker recomputes the "Expires in …" row label every 30s. Runs until process exit.
+// runSessionExpiryTicker recomputes the "Expires in …" row label until process exit.
+// The interval scales with the remaining time: coarse when the deadline is far off,
+// down to 10s in the final two minutes so the label doesn't lag the ceiling-rounded
+// countdown near expiry. The cached deadline is re-read every iteration, so an extend
+// or reconnect that moves it is picked up on the next tick.
 func (t *Tray) runSessionExpiryTicker() {
-	tk := time.NewTicker(30 * time.Second)
-	for range tk.C {
+	tm := time.NewTimer(sessionRefreshInterval(t.sessionRemaining()))
+	defer tm.Stop()
+	for range tm.C {
 		t.refreshSessionExpiresLabel()
+		tm.Reset(sessionRefreshInterval(t.sessionRemaining()))
+	}
+}
+
+// sessionRemaining returns the time left on the cached SSO deadline, or 0 when unknown.
+func (t *Tray) sessionRemaining() time.Duration {
+	t.sessionMu.Lock()
+	deadline := t.sessionExpiresAt
+	t.sessionMu.Unlock()
+	if deadline.IsZero() {
+		return 0
+	}
+	return time.Until(deadline)
+}
+
+// sessionRefreshInterval picks how long to wait before the next label recompute.
+func sessionRefreshInterval(remaining time.Duration) time.Duration {
+	switch {
+	case remaining <= 0:
+		return 30 * time.Second
+	case remaining <= 2*time.Minute:
+		return 10 * time.Second
+	case remaining <= time.Hour:
+		return 30 * time.Second
+	default:
+		return time.Minute
 	}
 }
 
@@ -87,30 +118,39 @@ func (t *Tray) refreshSessionExpiresLabel() {
 	if deadline.IsZero() {
 		return
 	}
-	remaining := t.formatSessionRemaining(time.Until(deadline))
-	item.SetLabel(t.loc.T("tray.session.expiresIn", "remaining", remaining))
+	item.SetLabel(t.sessionRowLabel(deadline))
+}
+
+func (t *Tray) sessionRowLabel(deadline time.Time) string {
+	remaining := time.Until(deadline)
+	if remaining <= 0 {
+		return t.loc.T("tray.status.sessionExpired")
+	}
+	return t.loc.T("tray.session.expiresIn", "remaining", t.formatSessionRemaining(remaining))
 }
 
 // formatSessionRemaining renders d as a localised long-form string picking the largest non-zero unit.
+// Each unit is rounded up so the label never claims less time than actually remains, matching the
+// upper-bound sense of the sub-minute "less than a minute" fragment.
 // Singular/plural keys are split per language for proper translation.
 func (t *Tray) formatSessionRemaining(d time.Duration) string {
 	switch {
 	case d < time.Minute:
 		return t.loc.T("tray.session.unit.lessThanMinute")
-	case d < time.Hour:
-		m := int(d / time.Minute)
+	case d <= 59*time.Minute:
+		m := ceilDiv(d, time.Minute)
 		if m == 1 {
 			return t.loc.T("tray.session.unit.minute")
 		}
 		return t.loc.T("tray.session.unit.minutes", "count", strconv.Itoa(m))
-	case d < 24*time.Hour:
-		h := int((d + 30*time.Minute) / time.Hour)
+	case d <= 23*time.Hour:
+		h := ceilDiv(d, time.Hour)
 		if h == 1 {
 			return t.loc.T("tray.session.unit.hour")
 		}
 		return t.loc.T("tray.session.unit.hours", "count", strconv.Itoa(h))
 	default:
-		days := int((d + 12*time.Hour) / (24 * time.Hour))
+		days := ceilDiv(d, 24*time.Hour)
 		if days == 1 {
 			return t.loc.T("tray.session.unit.day")
 		}
@@ -118,6 +158,11 @@ func (t *Tray) formatSessionRemaining(d time.Duration) string {
 	}
 }
 
+// ceilDiv divides d by unit rounding up, assuming d > 0.
+func ceilDiv(d, unit time.Duration) int {
+	return int((d + unit - time.Nanosecond) / unit)
+}
+
 // registerSessionWarningCategory wires the OS notification category and response handler for the expiry warning.
 // Errors are swallowed since the worst case is a plain notification without buttons.
 func (t *Tray) registerSessionWarningCategory() {
@@ -252,11 +297,9 @@ func (t *Tray) openSessionExpiration() {
 }
 
 // openSessionExtendFlow opens the SessionExpiration window seeded with the cached deadline's remaining time,
-// for the "Expires in …" tray row. No-ops when the deadline is unknown or elapsed.
+// for the "Expires in …" tray row. Once the deadline has elapsed the row reads "Session expired" and the
+// click routes to the login flow instead. No-op when the deadline is unknown.
 func (t *Tray) openSessionExtendFlow() {
-	if t.svc.WindowManager == nil {
-		return
-	}
 	t.sessionMu.Lock()
 	deadline := t.sessionExpiresAt
 	t.sessionMu.Unlock()
@@ -265,6 +308,14 @@ func (t *Tray) openSessionExtendFlow() {
 	}
 	seconds := int(time.Until(deadline).Seconds())
 	if seconds <= 0 {
+		if t.window != nil {
+			t.window.SetURL("/#/login")
+			t.window.Show()
+			t.window.Focus()
+		}
+		return
+	}
+	if t.svc.WindowManager == nil {
 		return
 	}
 	t.svc.WindowManager.OpenSessionExpiration(seconds)

combined/cmd/admin.go

@@ -0,0 +1,151 @@
+package cmd
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/dexidp/dex/storage"
+	log "github.com/sirupsen/logrus"
+	"github.com/spf13/cobra"
+
+	"github.com/netbirdio/netbird/formatter/hook"
+	admincmd "github.com/netbirdio/netbird/management/cmd/admin"
+	tokencmd "github.com/netbirdio/netbird/management/cmd/token"
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
+	"github.com/netbirdio/netbird/management/server/activity"
+	activitystore "github.com/netbirdio/netbird/management/server/activity/store"
+	"github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/management/server/types"
+	"github.com/netbirdio/netbird/util"
+)
+
+// newAdminCommands creates the admin command tree with combined-specific resource openers.
+func newAdminCommands() *cobra.Command {
+	return admincmd.NewCommands(admincmd.Openers{
+		Resources: withAdminResources,
+		Store:     withAdminStoreOnly,
+		IDP:       withAdminIDPOnly,
+	})
+}
+
+func newLegacyTokenCommand() *cobra.Command {
+	cmd := tokencmd.NewCommands(tokencmd.StoreOpener(withAdminStoreOnly))
+	cmd.Deprecated = "use 'admin token' instead"
+	return cmd
+}
+
+// withAdminResources loads the combined YAML config, initializes stores, and calls fn.
+func withAdminResources(cmd *cobra.Command, fn func(ctx context.Context, resources admincmd.Resources) error) error {
+	return withAdminConfig(cmd, func(ctx context.Context, cfg *CombinedConfig) error {
+		mgmtConfig, err := adminManagementConfig(cfg)
+		if err != nil {
+			return err
+		}
+
+		managementStore, err := openAdminStore(ctx, cfg)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseStore(ctx, managementStore)
+
+		idpStorage, idpStorageFile, err := admincmd.OpenIDPStorage(mgmtConfig)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseIDPStorage(idpStorage)
+
+		eventStore, esErr := openAdminEventStore(ctx, cfg, mgmtConfig)
+		if esErr != nil {
+			_, _ = fmt.Fprintf(cmd.ErrOrStderr(), "Warning: audit events will not be recorded: %v\n", esErr)
+		}
+		if eventStore != nil {
+			defer func() {
+				if err := eventStore.Close(ctx); err != nil {
+					log.Debugf("close activity event store: %v", err)
+				}
+			}()
+		}
+
+		return fn(ctx, admincmd.Resources{Store: managementStore, IDPStorage: idpStorage, IDPStorageFile: idpStorageFile, EventStore: eventStore})
+	})
+}
+
+// withAdminStoreOnly opens only the management store for admin subcommands that do not
+// need embedded IdP storage.
+func withAdminStoreOnly(cmd *cobra.Command, fn func(ctx context.Context, s store.Store) error) error {
+	return withAdminConfig(cmd, func(ctx context.Context, cfg *CombinedConfig) error {
+		managementStore, err := openAdminStore(ctx, cfg)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseStore(ctx, managementStore)
+
+		return fn(ctx, managementStore)
+	})
+}
+
+func withAdminIDPOnly(cmd *cobra.Command, fn func(ctx context.Context, idpStorage storage.Storage, storageFile string) error) error {
+	return withAdminConfig(cmd, func(ctx context.Context, cfg *CombinedConfig) error {
+		mgmtConfig, err := adminManagementConfig(cfg)
+		if err != nil {
+			return err
+		}
+		idpStorage, idpStorageFile, err := admincmd.OpenIDPStorage(mgmtConfig)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseIDPStorage(idpStorage)
+
+		return fn(ctx, idpStorage, idpStorageFile)
+	})
+}
+
+func withAdminConfig(cmd *cobra.Command, fn func(ctx context.Context, cfg *CombinedConfig) error) error {
+	if err := util.InitLog("error", "console"); err != nil {
+		return fmt.Errorf("init log: %w", err)
+	}
+
+	ctx := context.WithValue(cmd.Context(), hook.ExecutionContextKey, hook.SystemSource) //nolint:staticcheck
+
+	cfg, err := LoadConfig(configPath)
+	if err != nil {
+		return fmt.Errorf("load config: %w", err)
+	}
+	cfg.ApplyAdminDefaults()
+	applyServerStoreEnv(cfg.Server.Store)
+
+	return fn(ctx, cfg)
+}
+
+func adminManagementConfig(cfg *CombinedConfig) (*nbconfig.Config, error) {
+	mgmtConfig, err := cfg.ToManagementConfig()
+	if err != nil {
+		return nil, fmt.Errorf("create management config: %w", err)
+	}
+	return mgmtConfig, nil
+}
+
+func openAdminStore(ctx context.Context, cfg *CombinedConfig) (store.Store, error) {
+	managementStore, err := store.NewStore(ctx, types.Engine(cfg.Management.Store.Engine), cfg.Management.DataDir, nil, true)
+	if err != nil {
+		return nil, fmt.Errorf("create store: %w", err)
+	}
+	return managementStore, nil
+}
+
+func openAdminEventStore(ctx context.Context, cfg *CombinedConfig, config *nbconfig.Config) (activity.Store, error) {
+	if config.DataStoreEncryptionKey == "" {
+		return nil, fmt.Errorf("data store encryption key is not configured")
+	}
+	if err := applyActivityStoreEnv(cfg.Server.ActivityStore); err != nil {
+		return nil, fmt.Errorf("configure activity event store: %w", err)
+	}
+	eventStore, err := activitystore.NewSqlStore(ctx, config.Datadir, config.DataStoreEncryptionKey)
+	if err != nil {
+		return nil, fmt.Errorf("open activity event store: %w", err)
+	}
+	if eventStore == nil {
+		return nil, fmt.Errorf("open activity event store: returned nil store")
+	}
+	return eventStore, nil
+}

combined/cmd/admin_config_test.go

@@ -0,0 +1,47 @@
+package cmd
+
+import (
+	"context"
+	"os"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
+)
+
+func TestApplyAdminDefaultsCopiesServerStoreWithoutExposedAddress(t *testing.T) {
+	cfg := DefaultConfig()
+	cfg.Server.ExposedAddress = ""
+	cfg.Server.DataDir = "/srv/netbird"
+	cfg.Server.Store = StoreConfig{
+		Engine: "postgres",
+		DSN:    "postgres://user:pass@example.com/netbird",
+	}
+
+	cfg.ApplyAdminDefaults()
+
+	require.Equal(t, "/srv/netbird", cfg.Management.DataDir)
+	require.Equal(t, "postgres", cfg.Management.Store.Engine)
+	require.Equal(t, cfg.Server.Store.DSN, cfg.Management.Store.DSN)
+}
+
+func TestOpenAdminEventStoreMissingEncryptionKeyReturnsNilInterface(t *testing.T) {
+	eventStore, err := openAdminEventStore(context.Background(), &CombinedConfig{}, &nbconfig.Config{})
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "encryption key")
+	require.Nil(t, eventStore)
+}
+
+func TestApplyServerStoreEnv(t *testing.T) {
+	t.Setenv("NB_STORE_ENGINE_POSTGRES_DSN", "")
+	t.Setenv("NB_STORE_ENGINE_MYSQL_DSN", "")
+	t.Setenv("NB_STORE_ENGINE_SQLITE_FILE", "")
+
+	applyServerStoreEnv(StoreConfig{Engine: "postgres", DSN: "postgres-dsn", File: "store.db"})
+	require.Equal(t, "postgres-dsn", os.Getenv("NB_STORE_ENGINE_POSTGRES_DSN"))
+	require.Equal(t, "store.db", os.Getenv("NB_STORE_ENGINE_SQLITE_FILE"))
+
+	applyServerStoreEnv(StoreConfig{Engine: "mysql", DSN: "mysql-dsn"})
+	require.Equal(t, "mysql-dsn", os.Getenv("NB_STORE_ENGINE_MYSQL_DSN"))
+}

combined/cmd/config.go

@@ -6,8 +6,7 @@ import (
 	"net"
 	"net/netip"
 	"os"
-	"path"
-	"path/filepath"
+	filePath "path/filepath"
 	"strings"
 	"time"
 
@@ -74,6 +73,9 @@ type ServerConfig struct {
 	ActivityStore           StoreConfig        `yaml:"activityStore"`
 	AuthStore               StoreConfig        `yaml:"authStore"`
 	ReverseProxy            ReverseProxyConfig `yaml:"reverseProxy"`
+
+	SupportedSyncMessageVersions           *int           `yaml:"supportedSyncMessageVersions,omitempty"`
+	PerAccountSupportedSyncMessageVersions map[string]int `yaml:"perAccountSupportedSyncMessageVersions,omitempty"`
 }
 
 // TLSConfig contains TLS/HTTPS settings
@@ -300,6 +302,19 @@ func (c *CombinedConfig) ApplySimplifiedDefaults() {
 	c.autoConfigureClientSettings(exposedProto, exposedHost, exposedHostPort, hasExternalStuns, hasExternalRelay, hasExternalSignal)
 }
 
+// ApplyAdminDefaults applies the management settings needed by admin commands even
+// when the full server config is invalid and ApplySimplifiedDefaults cannot run.
+func (c *CombinedConfig) ApplyAdminDefaults() {
+	if c.Management.DataDir == "" || c.Management.DataDir == "/var/lib/netbird/" {
+		c.Management.DataDir = c.Server.DataDir
+	}
+	if c.Management.Store.Engine == "" || c.Management.Store.Engine == "sqlite" {
+		if c.Server.Store.Engine != "" || c.Server.Store.File != "" || c.Server.Store.DSN != "" {
+			c.Management.Store = c.Server.Store
+		}
+	}
+}
+
 // applyRelayDefaults configures the relay service if no external relay is configured.
 func (c *CombinedConfig) applyRelayDefaults(exposedProto, exposedHostPort string, hasExternalRelay, hasExternalStuns bool) {
 	if hasExternalRelay {
@@ -577,11 +592,11 @@ func (c *CombinedConfig) buildEmbeddedIdPConfig(mgmt ManagementConfig) (*idp.Emb
 			return nil, fmt.Errorf("authStore.dsn is required when authStore.engine is postgres")
 		}
 	} else {
-		authStorageFile = path.Join(mgmt.DataDir, "idp.db")
+		authStorageFile = filePath.Join(mgmt.DataDir, "idp.db")
 		if c.Server.AuthStore.File != "" {
 			authStorageFile = c.Server.AuthStore.File
-			if !filepath.IsAbs(authStorageFile) {
-				authStorageFile = filepath.Join(mgmt.DataDir, authStorageFile)
+			if !filePath.IsAbs(authStorageFile) {
+				authStorageFile = filePath.Join(mgmt.DataDir, authStorageFile)
 			}
 		}
 	}
@@ -696,16 +711,18 @@ func (c *CombinedConfig) ToManagementConfig() (*nbconfig.Config, error) {
 	httpConfig.AuthCallbackURL = callbackURL + types.ProxyCallbackEndpointFull
 
 	return &nbconfig.Config{
-		Stuns:                  stuns,
-		Relay:                  relayConfig,
-		Signal:                 signalConfig,
-		Datadir:                mgmt.DataDir,
-		DataStoreEncryptionKey: mgmt.Store.EncryptionKey,
-		HttpConfig:             httpConfig,
-		StoreConfig:            storeConfig,
-		ReverseProxy:           reverseProxy,
-		DisableDefaultPolicy:   mgmt.DisableDefaultPolicy,
-		EmbeddedIdP:            embeddedIdP,
+		Stuns:                              stuns,
+		Relay:                              relayConfig,
+		Signal:                             signalConfig,
+		Datadir:                            mgmt.DataDir,
+		DataStoreEncryptionKey:             mgmt.Store.EncryptionKey,
+		HttpConfig:                         httpConfig,
+		StoreConfig:                        storeConfig,
+		ReverseProxy:                       reverseProxy,
+		DisableDefaultPolicy:               mgmt.DisableDefaultPolicy,
+		EmbeddedIdP:                        embeddedIdP,
+		HighestSupportedSyncMessageVersion: c.Server.SupportedSyncMessageVersions,
+		PerAccountHighestSupportedSyncMessageVersion: c.Server.PerAccountSupportedSyncMessageVersions,
 	}, nil
 }
 
@@ -729,7 +746,7 @@ func ApplyEmbeddedIdPConfig(ctx context.Context, cfg *nbconfig.Config, mgmtPort
 		cfg.EmbeddedIdP.Storage.Type = "sqlite3"
 	}
 	if cfg.EmbeddedIdP.Storage.Config.File == "" && cfg.Datadir != "" {
-		cfg.EmbeddedIdP.Storage.Config.File = path.Join(cfg.Datadir, "idp.db")
+		cfg.EmbeddedIdP.Storage.Config.File = filePath.Join(cfg.Datadir, "idp.db")
 	}
 
 	issuer := cfg.EmbeddedIdP.Issuer

combined/cmd/root.go

@@ -31,6 +31,7 @@ import (
 	relayServer "github.com/netbirdio/netbird/relay/server"
 	"github.com/netbirdio/netbird/relay/server/listener"
 	"github.com/netbirdio/netbird/relay/server/listener/ws"
+	syncgrpc "github.com/netbirdio/netbird/shared/management/grpc"
 	sharedMetrics "github.com/netbirdio/netbird/shared/metrics"
 	"github.com/netbirdio/netbird/shared/relay/auth"
 	"github.com/netbirdio/netbird/shared/signal/proto"
@@ -64,7 +65,8 @@ func init() {
 	rootCmd.PersistentFlags().StringVarP(&configPath, "config", "c", "", "path to YAML configuration file (required)")
 	_ = rootCmd.MarkPersistentFlagRequired("config")
 
-	rootCmd.AddCommand(newTokenCommands())
+	rootCmd.AddCommand(newAdminCommands())
+	rootCmd.AddCommand(newLegacyTokenCommand())
 }
 
 func RootCmd() *cobra.Command {
@@ -122,6 +124,37 @@ func execute(cmd *cobra.Command, _ []string) error {
 }
 
 // initializeConfig loads and validates the configuration, then initializes logging.
+func applyServerStoreEnv(storeConfig StoreConfig) {
+	if dsn := storeConfig.DSN; dsn != "" {
+		switch strings.ToLower(storeConfig.Engine) {
+		case "postgres":
+			os.Setenv("NB_STORE_ENGINE_POSTGRES_DSN", dsn)
+		case "mysql":
+			os.Setenv("NB_STORE_ENGINE_MYSQL_DSN", dsn)
+		}
+	}
+	if file := storeConfig.File; file != "" {
+		os.Setenv("NB_STORE_ENGINE_SQLITE_FILE", file)
+	}
+}
+
+func applyActivityStoreEnv(storeConfig StoreConfig) error {
+	if engine := storeConfig.Engine; engine != "" {
+		engineLower := strings.ToLower(engine)
+		if engineLower == "postgres" && storeConfig.DSN == "" {
+			return fmt.Errorf("activityStore.dsn is required when activityStore.engine is postgres")
+		}
+		os.Setenv("NB_ACTIVITY_EVENT_STORE_ENGINE", engineLower)
+		if dsn := storeConfig.DSN; dsn != "" {
+			os.Setenv("NB_ACTIVITY_EVENT_POSTGRES_DSN", dsn)
+		}
+	}
+	if file := storeConfig.File; file != "" {
+		os.Setenv("NB_ACTIVITY_EVENT_SQLITE_FILE", file)
+	}
+	return nil
+}
+
 func initializeConfig() error {
 	var err error
 	config, err = LoadConfig(configPath)
@@ -137,30 +170,10 @@ func initializeConfig() error {
 		return fmt.Errorf("failed to initialize log: %w", err)
 	}
 
-	if dsn := config.Server.Store.DSN; dsn != "" {
-		switch strings.ToLower(config.Server.Store.Engine) {
-		case "postgres":
-			os.Setenv("NB_STORE_ENGINE_POSTGRES_DSN", dsn)
-		case "mysql":
-			os.Setenv("NB_STORE_ENGINE_MYSQL_DSN", dsn)
-		}
-	}
-	if file := config.Server.Store.File; file != "" {
-		os.Setenv("NB_STORE_ENGINE_SQLITE_FILE", file)
-	}
+	applyServerStoreEnv(config.Server.Store)
 
-	if engine := config.Server.ActivityStore.Engine; engine != "" {
-		engineLower := strings.ToLower(engine)
-		if engineLower == "postgres" && config.Server.ActivityStore.DSN == "" {
-			return fmt.Errorf("activityStore.dsn is required when activityStore.engine is postgres")
-		}
-		os.Setenv("NB_ACTIVITY_EVENT_STORE_ENGINE", engineLower)
-		if dsn := config.Server.ActivityStore.DSN; dsn != "" {
-			os.Setenv("NB_ACTIVITY_EVENT_POSTGRES_DSN", dsn)
-		}
-	}
-	if file := config.Server.ActivityStore.File; file != "" {
-		os.Setenv("NB_ACTIVITY_EVENT_SQLITE_FILE", file)
+	if err := applyActivityStoreEnv(config.Server.ActivityStore); err != nil {
+		return err
 	}
 
 	log.Infof("Starting combined NetBird server")
@@ -505,6 +518,16 @@ func createManagementServer(cfg *CombinedConfig, mgmtConfig *nbconfig.Config) (m
 	}
 	mgmtPort, _ := strconv.Atoi(portStr)
 
+	if err := syncgrpc.ValidateSyncMessageVersion(mgmtConfig.HighestSupportedSyncMessageVersion); err != nil {
+		return nil, err
+	}
+
+	for accountId, version := range mgmtConfig.PerAccountHighestSupportedSyncMessageVersion {
+		if err := syncgrpc.ValidateSyncMessageVersion(&version); err != nil {
+			return nil, fmt.Errorf("unrecognized sync message version in perAccountSupportedSyncMessageVersions for account %s %w", accountId, err)
+		}
+	}
+
 	mgmtSrv := newServer(
 		&mgmtServer.Config{
 			NbConfig:                mgmtConfig,

combined/cmd/token.go

@@ -1,63 +0,0 @@
-package cmd
-
-import (
-	"context"
-	"fmt"
-	"os"
-	"strings"
-
-	log "github.com/sirupsen/logrus"
-	"github.com/spf13/cobra"
-
-	"github.com/netbirdio/netbird/formatter/hook"
-	tokencmd "github.com/netbirdio/netbird/management/cmd/token"
-	"github.com/netbirdio/netbird/management/server/store"
-	"github.com/netbirdio/netbird/management/server/types"
-	"github.com/netbirdio/netbird/util"
-)
-
-// newTokenCommands creates the token command tree with combined-specific store opener.
-func newTokenCommands() *cobra.Command {
-	return tokencmd.NewCommands(withTokenStore)
-}
-
-// withTokenStore loads the combined YAML config, initializes the store, and calls fn.
-func withTokenStore(cmd *cobra.Command, fn func(ctx context.Context, s store.Store) error) error {
-	if err := util.InitLog("error", "console"); err != nil {
-		return fmt.Errorf("init log: %w", err)
-	}
-
-	ctx := context.WithValue(cmd.Context(), hook.ExecutionContextKey, hook.SystemSource) //nolint:staticcheck
-
-	cfg, err := LoadConfig(configPath)
-	if err != nil {
-		return fmt.Errorf("load config: %w", err)
-	}
-
-	if dsn := cfg.Server.Store.DSN; dsn != "" {
-		switch strings.ToLower(cfg.Server.Store.Engine) {
-		case "postgres":
-			os.Setenv("NB_STORE_ENGINE_POSTGRES_DSN", dsn)
-		case "mysql":
-			os.Setenv("NB_STORE_ENGINE_MYSQL_DSN", dsn)
-		}
-	}
-	if file := cfg.Server.Store.File; file != "" {
-		os.Setenv("NB_STORE_ENGINE_SQLITE_FILE", file)
-	}
-
-	datadir := cfg.Management.DataDir
-	engine := types.Engine(cfg.Management.Store.Engine)
-
-	s, err := store.NewStore(ctx, engine, datadir, nil, true)
-	if err != nil {
-		return fmt.Errorf("create store: %w", err)
-	}
-	defer func() {
-		if err := s.Close(ctx); err != nil {
-			log.Debugf("close store: %v", err)
-		}
-	}()
-
-	return fn(ctx, s)
-}

dns/nameserver.go

@@ -53,6 +53,7 @@ type NameServerGroup struct {
 	ID string `gorm:"primaryKey"`
 	// AccountID is a reference to Account that this object belongs
 	AccountID string `gorm:"index"`
+	PublicID  string `json:"-"`
 	// Name group name
 	Name string
 	// Description group description

docs/io.netbird.client.plist

@@ -66,6 +66,9 @@
     <key>disableAutoConnect</key>
     <false/>
 
+    <key>disableAutostart</key>
+    <false/>
+
     <key>disableClientRoutes</key>
     <false/>
 

docs/netbird-macos.mobileconfig

@@ -103,6 +103,8 @@
                                 <!--
                                 <key>disableAutoConnect</key>
                                 <false/>
+                                <key>disableAutostart</key>
+                                <false/>
                                 <key>disableClientRoutes</key>
                                 <false/>
                                 <key>disableServerRoutes</key>

docs/netbird-macos.sh

@@ -58,6 +58,7 @@ preSharedKey="$NULL"                       # secret; redacted in log
 allowServerSSH='true'
 blockInbound="$NULL"
 disableAutoConnect="$NULL"
+disableAutostart="$NULL"
 disableClientRoutes="$NULL"
 disableServerRoutes="$NULL"
 disableMetricsCollection="$NULL"
@@ -155,6 +156,7 @@ main() {
   is_set "$allowServerSSH"            && emit_bool    allowServerSSH            "$allowServerSSH"
   is_set "$blockInbound"              && emit_bool    blockInbound              "$blockInbound"
   is_set "$disableAutoConnect"        && emit_bool    disableAutoConnect        "$disableAutoConnect"
+  is_set "$disableAutostart"          && emit_bool    disableAutostart          "$disableAutostart"
   is_set "$disableClientRoutes"       && emit_bool    disableClientRoutes       "$disableClientRoutes"
   is_set "$disableServerRoutes"       && emit_bool    disableServerRoutes       "$disableServerRoutes"
   is_set "$disableMetricsCollection"  && emit_bool    disableMetricsCollection  "$disableMetricsCollection"

docs/netbird.adml

@@ -24,6 +24,9 @@
       <string id="DisableAutoConnect_Name">Disable auto-connect</string>
       <string id="DisableAutoConnect_Help">When enabled, the NetBird tunnel does not auto-connect at daemon startup. Equivalent to --disable-auto-connect.</string>
 
+      <string id="DisableAutostart_Name">Disable autostart</string>
+      <string id="DisableAutostart_Help">When enabled, the NetBird GUI is prevented from registering itself as an OS autostart entry on fresh installs, and any existing OS autostart entry registration is removed on the next GUI launch (Windows Registry Run key, macOS Login Item, Linux .desktop). Once the admin lifts the policy, the setting stays off until the user re-enables it in Settings.</string>
+
       <string id="DisableClientRoutes_Name">Disable client routes</string>
       <string id="DisableClientRoutes_Help">When enabled, this client will not consume routes advertised by routing peers. Equivalent to --disable-client-routes.</string>
 

docs/netbird.admx

@@ -64,6 +64,18 @@
       <disabledValue><decimal value="0" /></disabledValue>
     </policy>
 
+    <policy name="DisableAutostart"
+            class="Machine"
+            displayName="$(string.DisableAutostart_Name)"
+            explainText="$(string.DisableAutostart_Help)"
+            key="Software\Policies\NetBird"
+            valueName="DisableAutostart">
+      <parentCategory ref="NetBird" />
+      <supportedOn ref="SUPPORTED_NetBird_All" />
+      <enabledValue><decimal value="1" /></enabledValue>
+      <disabledValue><decimal value="0" /></disabledValue>
+    </policy>
+
     <policy name="DisableClientRoutes"
             class="Machine"
             displayName="$(string.DisableClientRoutes_Name)"

e2e/agentnetwork/chat_test.go

@@ -9,25 +9,234 @@ import (
 	"testing"
 	"time"
 
+	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"gorm.io/driver/sqlite"
+	"gorm.io/gorm"
 
 	"github.com/netbirdio/netbird/e2e/harness"
 	"github.com/netbirdio/netbird/shared/management/http/api"
 )
 
+// per1k is a model's published USD rates per 1k tokens. read is the prompt-cache read rate
+// (OpenAI: the cached-input discount rate); write is the cache-creation rate where one exists.
+type per1k struct{ in, out, read, write float64 }
+
+// publishedPer1k hardcodes the vendors' PUBLISHED rates for the models the live matrix can drive,
+// keyed by the normalized model id the proxy stamps. Deliberately independent of the proxy's
+// pricing table so a wrong embedded rate or a broken normalization fails the run.
+var publishedPer1k = map[string]per1k{
+	"gpt-4o-mini":                 {0.00015, 0.0006, 0.000075, 0},
+	"gpt-4o":                      {0.0025, 0.01, 0.00125, 0},
+	"claude-haiku-4-5":            {0.001, 0.005, 0.0001, 0.00125},
+	"claude-sonnet-4-5":           {0.003, 0.015, 0.0003, 0.00375},
+	"claude-sonnet-4-6":           {0.003, 0.015, 0.0003, 0.00375},
+	"kimi-k3":                     {0.003, 0.015, 0.0003, 0.003}, // no published write rate: bills at the input rate
+	"anthropic.claude-haiku-4-5":  {0.001, 0.005, 0.0001, 0.00125},
+	"anthropic.claude-sonnet-4-5": {0.003, 0.015, 0.0003, 0.00375},
+	"anthropic.claude-sonnet-4-6": {0.003, 0.015, 0.0003, 0.00375},
+}
+
+// rawCostVerificationSQL is the operator-facing double-check, run straight against the management
+// sqlite store: recompute each usage row's expected total and cache cost from its own persisted
+// token buckets and hardcoded published rates. OpenAI counts cached tokens as a subset of input;
+// Anthropic-shape providers count cache buckets additively.
+const rawCostVerificationSQL = `
+WITH rates(model, in_rate, out_rate, read_rate, write_rate) AS (
+  VALUES
+    ('gpt-4o-mini',                 0.00015, 0.0006, 0.000075, 0.0),
+    ('gpt-4o',                      0.0025,  0.01,   0.00125,  0.0),
+    ('claude-haiku-4-5',            0.001,   0.005,  0.0001,   0.00125),
+    ('claude-sonnet-4-5',           0.003,   0.015,  0.0003,   0.00375),
+    ('claude-sonnet-4-6',           0.003,   0.015,  0.0003,   0.00375),
+    ('kimi-k3',                     0.003,   0.015,  0.0003,   0.003),
+    ('anthropic.claude-haiku-4-5',  0.001,   0.005,  0.0001,   0.00125),
+    ('anthropic.claude-sonnet-4-5', 0.003,   0.015,  0.0003,   0.00375),
+    ('anthropic.claude-sonnet-4-6', 0.003,   0.015,  0.0003,   0.00375)
+)
+SELECT
+  u.provider,
+  u.model,
+  u.input_tokens,
+  u.output_tokens,
+  u.cached_input_tokens,
+  u.cache_creation_tokens,
+  u.input_cost_usd,
+  u.cached_input_cost_usd,
+  u.cache_creation_cost_usd,
+  u.output_cost_usd,
+  -- No cost_usd / cache_cost_usd columns are stored: both are derived from the
+  -- four per-bucket columns above, exactly as the API renders them.
+  (u.input_cost_usd + u.cached_input_cost_usd + u.cache_creation_cost_usd + u.output_cost_usd) AS cost_usd,
+  (u.cached_input_cost_usd + u.cache_creation_cost_usd) AS cache_cost_usd,
+  CASE WHEN u.provider = 'openai' THEN
+    (u.input_tokens - MIN(u.cached_input_tokens, u.input_tokens))*r.in_rate/1000.0
+  ELSE
+    u.input_tokens*r.in_rate/1000.0
+  END AS expected_input,
+  CASE WHEN u.provider = 'openai' THEN
+    MIN(u.cached_input_tokens, u.input_tokens)*r.read_rate/1000.0
+  ELSE
+    u.cached_input_tokens*r.read_rate/1000.0
+  END AS expected_cached_input,
+  CASE WHEN u.provider = 'openai' THEN
+    0.0
+  ELSE
+    u.cache_creation_tokens*r.write_rate/1000.0
+  END AS expected_cache_creation,
+  u.output_tokens*r.out_rate/1000.0 AS expected_output,
+  CASE WHEN u.provider = 'openai' THEN
+    (u.input_tokens - MIN(u.cached_input_tokens, u.input_tokens))*r.in_rate/1000.0
+      + MIN(u.cached_input_tokens, u.input_tokens)*r.read_rate/1000.0
+      + u.output_tokens*r.out_rate/1000.0
+  ELSE
+    u.input_tokens*r.in_rate/1000.0 + u.cached_input_tokens*r.read_rate/1000.0
+      + u.cache_creation_tokens*r.write_rate/1000.0 + u.output_tokens*r.out_rate/1000.0
+  END AS expected_total,
+  CASE WHEN u.provider = 'openai' THEN
+    MIN(u.cached_input_tokens, u.input_tokens)*r.read_rate/1000.0
+  ELSE
+    u.cached_input_tokens*r.read_rate/1000.0 + u.cache_creation_tokens*r.write_rate/1000.0
+  END AS expected_cache
+FROM agent_network_request_usage u
+JOIN rates r ON r.model = u.model
+ORDER BY u.timestamp`
+
+// verifyUsageRowsSQL re-checks every persisted usage row directly in the management sqlite store,
+// bypassing the API path — the same audit an operator can run on a production store.db.
+func verifyUsageRowsSQL(t *testing.T, srv *harness.Combined) {
+	t.Helper()
+
+	dbPath, err := srv.SnapshotStoreDB(t.TempDir())
+	require.NoError(t, err, "snapshot management sqlite store")
+
+	db, err := gorm.Open(sqlite.Open(dbPath), &gorm.Config{})
+	require.NoError(t, err, "open store snapshot")
+	sqlDB, err := db.DB()
+	require.NoError(t, err)
+	defer func() { _ = sqlDB.Close() }()
+
+	rows, err := db.Raw(rawCostVerificationSQL).Rows()
+	require.NoError(t, err, "run raw cost verification query")
+	defer func() { _ = rows.Close() }()
+
+	verified := 0
+	for rows.Next() {
+		var provider, model string
+		var inTok, outTok, readTok, writeTok int64
+		var inCost, cachedInCost, cacheCreateCost, outCost, cost, cacheCost float64
+		var wantInput, wantCachedInput, wantCacheCreation, wantOutput, wantTotal, wantCache float64
+		require.NoError(t, rows.Scan(&provider, &model, &inTok, &outTok, &readTok, &writeTok,
+			&inCost, &cachedInCost, &cacheCreateCost, &outCost, &cost, &cacheCost,
+			&wantInput, &wantCachedInput, &wantCacheCreation, &wantOutput, &wantTotal, &wantCache), "scan usage row")
+		t.Logf("[sql] %s/%s: in=%d out=%d cache_read=%d cache_write=%d stored in/cached/create/out=$%.6f/$%.6f/$%.6f/$%.6f total=$%.6f cache=$%.6f expected total=$%.6f cache=$%.6f",
+			provider, model, inTok, outTok, readTok, writeTok,
+			inCost, cachedInCost, cacheCreateCost, outCost, cost, cacheCost, wantTotal, wantCache)
+		assert.InDeltaf(t, wantInput, inCost, 1e-6, "stored input_cost_usd for %s/%s must match the published-rate recompute", provider, model)
+		assert.InDeltaf(t, wantCachedInput, cachedInCost, 1e-6, "stored cached_input_cost_usd for %s/%s must match the published-rate recompute", provider, model)
+		assert.InDeltaf(t, wantCacheCreation, cacheCreateCost, 1e-6, "stored cache_creation_cost_usd for %s/%s must match the published-rate recompute", provider, model)
+		assert.InDeltaf(t, wantOutput, outCost, 1e-6, "stored output_cost_usd for %s/%s must match the published-rate recompute", provider, model)
+		assert.InDeltaf(t, wantTotal, cost, 1e-6, "derived cost_usd for %s/%s must match the published-rate recompute", provider, model)
+		assert.InDeltaf(t, wantCache, cacheCost, 1e-6, "derived cache_cost_usd for %s/%s must match the published-rate recompute", provider, model)
+		assert.InDeltaf(t, inCost+cachedInCost+cacheCreateCost+outCost, cost, 1e-9,
+			"stored buckets must sum to the derived cost_usd for %s/%s", provider, model)
+		verified++
+	}
+	require.NoError(t, rows.Err(), "iterate usage rows")
+	require.Positive(t, verified, "raw SQL check must cover at least one usage row")
+	t.Logf("[sql] verified %d usage rows in store.db against published rates", verified)
+
+	gwRows, err := db.Raw(`SELECT model,
+	  (input_cost_usd + cached_input_cost_usd + cache_creation_cost_usd + output_cost_usd) AS cost_usd
+	  FROM agent_network_request_usage WHERE model LIKE '%/%'`).Rows()
+	require.NoError(t, err, "query gateway-prefixed usage rows")
+	defer func() { _ = gwRows.Close() }()
+	for gwRows.Next() {
+		var model string
+		var cost float64
+		require.NoError(t, gwRows.Scan(&model, &cost), "scan gateway usage row")
+		t.Logf("[sql] gateway %s: stored=$%.6f (must be 0 — deliberately unpriced)", model, cost)
+		assert.Zerof(t, cost, "gateway-prefixed model %q must store cost 0, never a guessed rate", model)
+	}
+	require.NoError(t, gwRows.Err(), "iterate gateway usage rows")
+}
+
+// validateAccessLogCost recomputes a live access-log row's expected total and cache cost from the
+// published per-1k rates and the row's persisted token buckets, and asserts both stored values.
+// Gateway-prefixed model ids the proxy deliberately does not price must store cost 0.
+func validateAccessLogCost(t *testing.T, pc providerCase, row api.AgentNetworkAccessLog) {
+	t.Helper()
+	model := catalogModel(pc)
+	provider := ""
+	if row.Provider != nil {
+		provider = *row.Provider
+	}
+	t.Logf("[cost] %s: provider=%s model=%s in=%d out=%d total=%d cache_read=%d cache_write=%d cost=$%.6f cache_cost=$%.6f",
+		pc.name, provider, model, row.InputTokens, row.OutputTokens, row.TotalTokens,
+		row.CachedInputTokens, row.CacheCreationTokens, row.CostUsd, row.CacheCostUsd)
+
+	rates, known := publishedPer1k[model]
+	if !known {
+		if strings.Contains(model, "/") {
+			assert.Zerof(t, row.CostUsd, "gateway-prefixed model %q is not priced so the cost meter must skip (cost 0)", model)
+			return
+		}
+		t.Logf("[cost] %s: no published rate on file for model %q (env-overridden?); skipping cost validation", pc.name, model)
+		return
+	}
+
+	// input_tokens may legitimately be 0: Moonshot/Kimi reports fully cached prompts under the cache
+	// buckets only. Output and total must always be present on a priced row.
+	require.Positive(t, row.OutputTokens, "priced row must carry output tokens")
+	require.Positive(t, row.TotalTokens, "priced row must carry total tokens")
+
+	var wantInput, wantCachedInput, wantCacheCreation float64
+	if provider == "openai" {
+		cached := min(row.CachedInputTokens, row.InputTokens) // cached is a subset of input
+		wantInput = float64(row.InputTokens-cached) / 1000 * rates.in
+		wantCachedInput = float64(cached) / 1000 * rates.read
+		// OpenAI has no cache-write bucket; wantCacheCreation stays 0.
+	} else {
+		// Anthropic / Bedrock shape: cache buckets are additive to input_tokens.
+		wantInput = float64(row.InputTokens) / 1000 * rates.in
+		wantCachedInput = float64(row.CachedInputTokens) / 1000 * rates.read
+		wantCacheCreation = float64(row.CacheCreationTokens) / 1000 * rates.write
+	}
+	wantOutput := float64(row.OutputTokens) / 1000 * rates.out
+	wantCache := wantCachedInput + wantCacheCreation
+	wantTotal := wantInput + wantCache + wantOutput
+
+	t.Logf("[cost] %s: expecting input=$%.6f cached_input=$%.6f cache_creation=$%.6f output=$%.6f total=$%.6f cache=$%.6f from published rates",
+		pc.name, wantInput, wantCachedInput, wantCacheCreation, wantOutput, wantTotal, wantCache)
+	assert.InDeltaf(t, wantInput, row.InputCostUsd, 1e-6, "stored input_cost_usd for %s (%s)", pc.name, model)
+	assert.InDeltaf(t, wantCachedInput, row.CachedInputCostUsd, 1e-6, "stored cached_input_cost_usd for %s (%s)", pc.name, model)
+	assert.InDeltaf(t, wantCacheCreation, row.CacheCreationCostUsd, 1e-6, "stored cache_creation_cost_usd for %s (%s)", pc.name, model)
+	assert.InDeltaf(t, wantOutput, row.OutputCostUsd, 1e-6, "stored output_cost_usd for %s (%s)", pc.name, model)
+	assert.InDeltaf(t, wantTotal, row.CostUsd, 1e-6, "derived cost_usd for %s (%s)", pc.name, model)
+	assert.InDeltaf(t, wantCache, row.CacheCostUsd, 1e-6, "derived cache_cost_usd for %s (%s)", pc.name, model)
+
+	// The aggregates must be exactly the sum of the stored components, not an
+	// independently-computed figure that could drift from the breakdown.
+	assert.InDeltaf(t, row.InputCostUsd+row.CachedInputCostUsd+row.CacheCreationCostUsd+row.OutputCostUsd,
+		row.CostUsd, 1e-9, "stored buckets must sum to the derived cost_usd for %s (%s)", pc.name, model)
+	assert.InDeltaf(t, row.CachedInputCostUsd+row.CacheCreationCostUsd,
+		row.CacheCostUsd, 1e-9, "stored cache buckets must sum to the derived cache_cost_usd for %s (%s)", pc.name, model)
+}
+
 // providerCase is one entry in the live provider matrix. The same scenario runs
 // for every available provider; availability is keyed off env vars so the suite
 // covers whatever credentials are present (source ~/.llm-keys locally / set the
 // Actions secrets in CI).
 type providerCase struct {
-	name      string
-	catalogID string
-	upstream  string
-	apiKey    string
-	model     string // body model (chat/messages) or path model@version (vertex)
-	kind      string // harness.WireChat, harness.WireMessages, or harness.WireVertex
-	project   string // vertex only: GCP project for the rawPredict path
-	region    string // vertex only: GCP region for the rawPredict path
+	name       string
+	catalogID  string
+	upstream   string
+	apiKey     string
+	model      string // body model (chat/messages) or path model@version (vertex)
+	kind       string // harness.WireChat, harness.WireMessages, or harness.WireVertex
+	project    string // vertex only: GCP project for the rawPredict path
+	region     string // vertex only: GCP region for the rawPredict path
+	pathPrefix string // base-URL path prefix the agent carries (e.g. "/anthropic" for Kimi)
 }
 
 // availableProviders builds the matrix from the provider env vars that are set.
@@ -39,6 +248,24 @@ func availableProviders() []providerCase {
 	if k := os.Getenv("ANTHROPIC_TOKEN"); k != "" {
 		ps = append(ps, providerCase{name: "anthropic", catalogID: "anthropic_api", upstream: "https://api.anthropic.com", apiKey: k, model: "claude-haiku-4-5", kind: harness.WireMessages})
 	}
+	if k := os.Getenv("KIMI_TOKEN"); k != "" {
+		// Kimi (Moonshot AI) serves two body shapes from the same key: OpenAI
+		// Chat Completions on the bare host (/v1/...) and the Anthropic
+		// Messages API under the /anthropic path prefix (the endpoint
+		// Moonshot's Claude Code guide uses). The provider keeps the bare
+		// default upstream and the AGENT carries the /anthropic prefix in
+		// its base URL — exactly the documented Claude Code / Kimi CLI
+		// setup (ANTHROPIC_BASE_URL=https://<endpoint>/anthropic) — so one
+		// provider serves both shapes and the prefix rides through to
+		// Moonshot. Run the Anthropic shape, the flagship Claude Code path;
+		// the OpenAI wire shape is covered live by the other chat-shaped
+		// matrix providers, and Kimi-over-chat passed with kimi-k3 before
+		// the single-model constraint surfaced (run #73 on the kimi feature
+		// branch). The platform serves this account exactly ONE model —
+		// kimi-k3 (kimi-k2-thinking and even kimi-latest return
+		// resource_not_found_error on both surfaces).
+		ps = append(ps, providerCase{name: "kimi", catalogID: "kimi_api", upstream: "https://api.moonshot.ai", apiKey: k, model: "kimi-k3", kind: harness.WireMessages, pathPrefix: "/anthropic"})
+	}
 	if k, u := os.Getenv("VERCEL_TOKEN"), os.Getenv("VERCEL_URL"); k != "" && u != "" {
 		ps = append(ps, providerCase{name: "vercel", catalogID: "vercel_ai_gateway", upstream: u, apiKey: k, model: "openai/gpt-4o-mini", kind: harness.WireChat})
 	}
@@ -84,14 +311,27 @@ func availableProviders() []providerCase {
 		}
 	}
 
-	// Bedrock: path-routed, bearer auth. Model is a cross-region inference
-	// profile id (distinct string from the first-party Anthropic case).
+	// Bedrock: path-routed, bearer auth. Model is the FULL cross-region
+	// inference-profile id exactly as AWS issues it — region-family prefix
+	// plus the date/version suffix. A bare or wrong-region id makes Bedrock
+	// reject the request with "The provided model identifier is invalid"
+	// before any inference runs. The proxy normalizes this id to the catalog
+	// key (anthropic.claude-haiku-4-5) for routing/pricing/allowlists.
+	// Defaults pair eu-central-1 with the eu.* profile; AWS_REGION overrides
+	// the region and the prefix follows its family.
 	if k := os.Getenv("AWS_BEARER_TOKEN_BEDROCK"); k != "" {
 		region := os.Getenv("AWS_REGION")
 		if region == "" {
-			region = "us-east-1"
+			region = "eu-central-1"
 		}
-		ps = append(ps, providerCase{name: "bedrock", catalogID: "bedrock_api", upstream: "https://bedrock-runtime." + region + ".amazonaws.com", apiKey: k, model: "us.anthropic.claude-haiku-4-5", kind: harness.WireBedrock})
+		// A valid Bedrock inference-profile id, overridable per account (AWS_BEDROCK_MODEL, also the
+		// workflow's bedrock_model dispatch input). `global.` profiles work from any region. Defaults to
+		// Sonnet 4.6, whose id convention dropped the -YYYYMMDD-v1:0 suffix that Haiku 4.5 still carries.
+		model := os.Getenv("AWS_BEDROCK_MODEL")
+		if model == "" {
+			model = "global.anthropic.claude-sonnet-4-6"
+		}
+		ps = append(ps, providerCase{name: "bedrock", catalogID: "bedrock_api", upstream: "https://bedrock-runtime." + region + ".amazonaws.com", apiKey: k, model: model, kind: harness.WireBedrock})
 	}
 	return ps
 }
@@ -108,8 +348,16 @@ func providerRequest(pc providerCase) api.AgentNetworkProviderRequest {
 		Enabled:     ptr(true),
 	}
 	if pc.kind != harness.WireVertex {
+		// The router matches the normalized catalog id. Bedrock's request model
+		// travels as a region-prefixed inference-profile id in the URL path
+		// (us.anthropic...), which the router strips before matching, so register
+		// the normalized form here or routing fails as model_not_routable.
+		modelID := pc.model
+		if pc.kind == harness.WireBedrock {
+			modelID = catalogModel(pc)
+		}
 		req.Models = &[]api.AgentNetworkProviderModel{
-			{Id: pc.model, InputPer1k: 0.001, OutputPer1k: 0.002},
+			{Id: modelID, InputPer1k: 0.001, OutputPer1k: 0.002},
 		}
 	}
 	return req
@@ -201,11 +449,12 @@ func TestProvidersMatrix(t *testing.T) {
 	t.Cleanup(func() { _ = cl.Terminate(context.Background()) })
 
 	require.NoError(t, cl.WaitConnected(ctx, 90*time.Second), "client must connect to management")
+	// Probe first: the GET resolves the endpoint (DNS error fails) and its first packet wakes the lazy proxy peer, so WaitProxyPeer sees it connected; any HTTP status counts.
+	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
+	require.NoError(t, err, "resolve agent-network endpoint to proxy IP")
 	if err := cl.WaitProxyPeer(ctx, 180*time.Second); err != nil {
 		t.Fatalf("client did not see the proxy peer: %v\n=== proxy logs ===\n%s", err, px.Logs(context.Background()))
 	}
-	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
-	require.NoError(t, err, "resolve agent-network endpoint to proxy IP")
 
 	for _, pc := range matrix {
 		pc := pc
@@ -216,6 +465,10 @@ func TestProvidersMatrix(t *testing.T) {
 			// session id and confirm the marker propagated end-to-end.
 			sessionID := "e2e-session-" + pc.name
 
+			// A long-form prompt so completions carry realistic token counts for cost validation;
+			// max_tokens in the harness bodies (2048) lets the full answer through.
+			const matrixPrompt = "explain GitHub workflow in 1000 words"
+
 			// Retry briefly to absorb tunnel/DNS jitter on the first call.
 			var code int
 			var body string
@@ -226,11 +479,11 @@ func TestProvidersMatrix(t *testing.T) {
 				var cerr error
 				switch pc.kind {
 				case harness.WireVertex:
-					c, b, cerr = cl.Vertex(ctx, settings.Endpoint, proxyIP, pc.project, pc.region, pc.model, "Reply with exactly: pong", sessionID)
+					c, b, cerr = cl.Vertex(ctx, settings.Endpoint, proxyIP, pc.project, pc.region, pc.model, matrixPrompt, sessionID)
 				case harness.WireBedrock:
-					c, b, cerr = cl.Bedrock(ctx, settings.Endpoint, proxyIP, pc.model, "Reply with exactly: pong", sessionID)
+					c, b, cerr = cl.Bedrock(ctx, settings.Endpoint, proxyIP, pc.model, matrixPrompt, sessionID)
 				default:
-					c, b, cerr = cl.Chat(ctx, settings.Endpoint, proxyIP, pc.kind, pc.model, "Reply with exactly: pong", sessionID)
+					c, b, cerr = cl.ChatPrefixed(ctx, settings.Endpoint, proxyIP, pc.pathPrefix, pc.kind, pc.model, matrixPrompt, sessionID)
 				}
 				if cerr == nil {
 					code, body = c, b
@@ -249,6 +502,7 @@ func TestProvidersMatrix(t *testing.T) {
 
 			// The session id sent as x-session-id must round-trip into the
 			// access-log row for this provider.
+			var row api.AgentNetworkAccessLog
 			require.Eventually(t, func() bool {
 				logs, lerr := srv.ListAccessLogs(ctx)
 				if lerr != nil {
@@ -256,11 +510,15 @@ func TestProvidersMatrix(t *testing.T) {
 				}
 				for _, r := range logs.Data {
 					if r.SessionId != nil && *r.SessionId == sessionID {
+						row = r
 						return true
 					}
 				}
 				return false
 			}, 30*time.Second, 2*time.Second, "session id %q must be recorded in an access-log row for %s", sessionID, pc.name)
+
+			// Stored total and cache cost must match the published rates applied to the row's buckets.
+			validateAccessLogCost(t, pc, row)
 		})
 	}
 
@@ -281,4 +539,7 @@ func TestProvidersMatrix(t *testing.T) {
 		}
 		return false
 	}, 60*time.Second, 3*time.Second, "consumption must be recorded with positive token counts after live traffic")
+
+	// Final raw-SQL audit: bypass the API and re-verify every persisted usage row in the store.
+	verifyUsageRowsSQL(t, srv)
 }

e2e/agentnetwork/guardrail_test.go

@@ -4,6 +4,7 @@ package agentnetwork
 
 import (
 	"context"
+	"regexp"
 	"strings"
 	"testing"
 	"time"
@@ -15,13 +16,29 @@ import (
 	"github.com/netbirdio/netbird/shared/management/http/api"
 )
 
+// bedrockRegionPrefixes and bedrockVersionSuffix mirror the proxy's Bedrock
+// model normalization (region/inference-profile prefix + version suffix) so the
+// provider is registered under the same catalog key the router matches against.
+var (
+	bedrockRegionPrefixes = []string{"us.", "eu.", "apac.", "global."}
+	bedrockVersionSuffix  = regexp.MustCompile(`-(\d{8}-)?v\d+(:\d+)?$`)
+)
+
 // catalogModel returns the normalized catalog id the proxy stamps for a
-// path-routed provider's configured model — the form the guardrail allowlist is
-// compared against (region prefix / @version stripped).
+// path-routed provider's configured model — the form the router and guardrail
+// allowlist compare against (Bedrock region prefix + version stripped, Vertex
+// @version stripped).
 func catalogModel(pc providerCase) string {
 	switch pc.kind {
 	case harness.WireBedrock:
-		return strings.TrimPrefix(pc.model, "us.")
+		m := pc.model
+		for _, p := range bedrockRegionPrefixes {
+			if strings.HasPrefix(m, p) {
+				m = m[len(p):]
+				break
+			}
+		}
+		return bedrockVersionSuffix.ReplaceAllString(m, "")
 	case harness.WireVertex:
 		return strings.SplitN(pc.model, "@", 2)[0]
 	default:
@@ -35,7 +52,9 @@ func catalogModel(pc providerCase) string {
 func disallowedModel(pc providerCase) string {
 	switch pc.kind {
 	case harness.WireBedrock:
-		return "us.anthropic.claude-opus-4-8"
+		// Same profile prefix as the allowed model so only the model name
+		// differs; the guardrail must deny it before it reaches AWS.
+		return strings.SplitN(pc.model, ".", 2)[0] + ".anthropic.claude-opus-4-8"
 	case harness.WireVertex:
 		return "claude-opus-4-8@20250101"
 	default:
@@ -55,7 +74,7 @@ func sendModel(ctx context.Context, t *testing.T, cl *harness.Client, endpoint,
 	case harness.WireVertex:
 		code, _, err = cl.Vertex(ctx, endpoint, proxyIP, pc.project, pc.region, model, "Reply with exactly: pong", "")
 	default:
-		code, _, err = cl.Chat(ctx, endpoint, proxyIP, pc.kind, model, "Reply with exactly: pong", "")
+		code, _, err = cl.ChatPrefixed(ctx, endpoint, proxyIP, pc.pathPrefix, pc.kind, model, "Reply with exactly: pong", "")
 	}
 	require.NoError(t, err, "request must reach the proxy for %s", pc.name)
 	return code
@@ -147,11 +166,12 @@ func TestModelAllowlistEnforced(t *testing.T) {
 	t.Cleanup(func() { _ = cl.Terminate(context.Background()) })
 
 	require.NoError(t, cl.WaitConnected(ctx, 90*time.Second), "client must connect to management")
+	// Probe first: the GET resolves the endpoint (DNS error fails) and its first packet wakes the lazy proxy peer, so WaitProxyPeer sees it connected; any HTTP status counts.
+	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
+	require.NoError(t, err, "resolve agent-network endpoint to proxy IP")
 	if err := cl.WaitProxyPeer(ctx, 180*time.Second); err != nil {
 		t.Fatalf("client did not see the proxy peer: %v\n=== proxy logs ===\n%s", err, px.Logs(context.Background()))
 	}
-	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
-	require.NoError(t, err, "resolve agent-network endpoint to proxy IP")
 
 	for _, pc := range providers {
 		pc := pc

e2e/agentnetwork/skiptls_test.go

@@ -104,11 +104,12 @@ func TestProviderSkipTLSVerification(t *testing.T) {
 	t.Cleanup(func() { _ = cl.Terminate(context.Background()) })
 
 	require.NoError(t, cl.WaitConnected(ctx, 90*time.Second), "client must connect to management")
+	// Probe first: the GET resolves the endpoint (DNS error fails) and its first packet wakes the lazy proxy peer, so WaitProxyPeer sees it connected; any HTTP status counts.
+	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
+	require.NoError(t, err, "resolve endpoint to proxy IP")
 	if err := cl.WaitProxyPeer(ctx, 180*time.Second); err != nil {
 		t.Fatalf("client did not see the proxy peer: %v\n=== proxy logs ===\n%s", err, px.Logs(context.Background()))
 	}
-	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
-	require.NoError(t, err, "resolve endpoint to proxy IP")
 
 	// Positive: skip=true reaches the self-signed upstream. Retry to absorb
 	// tunnel/DNS jitter on the first call; success also proves the path works.

e2e/agentnetwork/vllm_test.go

@@ -106,11 +106,12 @@ func TestVLLMProvider(t *testing.T) {
 	t.Cleanup(func() { _ = cl.Terminate(context.Background()) })
 
 	require.NoError(t, cl.WaitConnected(ctx, 90*time.Second), "client must connect to management")
+	// Probe first: the GET resolves the endpoint (DNS error fails) and its first packet wakes the lazy proxy peer, so WaitProxyPeer sees it connected; any HTTP status counts.
+	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
+	require.NoError(t, err, "resolve endpoint to proxy IP")
 	if err := cl.WaitProxyPeer(ctx, 180*time.Second); err != nil {
 		t.Fatalf("client did not see the proxy peer: %v\n=== proxy logs ===\n%s", err, px.Logs(context.Background()))
 	}
-	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
-	require.NoError(t, err, "resolve endpoint to proxy IP")
 
 	before, _ := srv.ListAccessLogs(ctx)
 	sessionID := "e2e-session-vllm"

e2e/harness/client.go

@@ -4,6 +4,7 @@ package harness
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"os/exec"
@@ -167,22 +168,54 @@ func (cl *Client) pollStatus(ctx context.Context, timeout time.Duration, want st
 	return fmt.Errorf("timed out waiting for %q; last status:\n%s", want, last)
 }
 
-// ResolveProxyIP resolves the agent-network endpoint to the proxy peer's
-// NetBird IP from inside the client (via magic DNS).
+const (
+	// curlExitCouldNotResolve is curl's exit code for a DNS resolution failure, distinct from connection-level failures.
+	curlExitCouldNotResolve = 6
+	// dnsProbeRetryWindow bounds DNS-failure retries: the synthesized zone lands a beat after management connects, so early NXDOMAIN is propagation; a zone still absent after this window is a real failure.
+	dnsProbeRetryWindow   = 30 * time.Second
+	dnsProbeRetryInterval = 2 * time.Second
+)
+
+// ResolveProxyIP GETs https://<endpoint>/ from the client's netns: any HTTP status proves DNS + tunnel and wakes the lazy proxy peer; only DNS failures retry, within dnsProbeRetryWindow. Returns the connected IP for --resolve pinning.
 func (cl *Client) ResolveProxyIP(ctx context.Context, endpoint string) (string, error) {
-	code, reader, err := cl.container.Exec(ctx, []string{"getent", "hosts", endpoint}, tcexec.Multiplexed())
-	if err != nil {
-		return "", err
+	args := []string{
+		"run", "--rm",
+		"--network", "container:" + cl.container.GetContainerID(),
+		curlImage,
+		"-ksS", "-o", "/dev/null",
+		"--connect-timeout", "30", "--max-time", "60",
+		"-w", "%{remote_ip}",
+		"https://" + endpoint + "/",
 	}
-	out, _ := io.ReadAll(reader)
-	if code != 0 {
-		return "", fmt.Errorf("getent hosts %s exited %d", endpoint, code)
+	deadline := time.Now().Add(dnsProbeRetryWindow)
+	for {
+		cmd := exec.CommandContext(ctx, "docker", args...)
+		var stdout, stderr strings.Builder
+		cmd.Stdout = &stdout
+		cmd.Stderr = &stderr
+		err := cmd.Run()
+		if err == nil {
+			ip := strings.TrimSpace(stdout.String())
+			if ip == "" {
+				return "", fmt.Errorf("got an HTTP response from %s but no remote IP", endpoint)
+			}
+			return ip, nil
+		}
+
+		var exitErr *exec.ExitError
+		if !errors.As(err, &exitErr) || exitErr.ExitCode() != curlExitCouldNotResolve {
+			return "", fmt.Errorf("no HTTP response from %s: %w (%s)", endpoint, err, strings.TrimSpace(stderr.String()))
+		}
+		dnsErr := fmt.Errorf("DNS resolution failed for %s: %s", endpoint, strings.TrimSpace(stderr.String()))
+		if time.Until(deadline) < dnsProbeRetryInterval {
+			return "", dnsErr
+		}
+		select {
+		case <-ctx.Done():
+			return "", fmt.Errorf("%w (%w)", dnsErr, ctx.Err())
+		case <-time.After(dnsProbeRetryInterval):
+		}
 	}
-	fields := strings.Fields(string(out))
-	if len(fields) == 0 {
-		return "", fmt.Errorf("no address for %s", endpoint)
-	}
-	return fields[0], nil
 }
 
 // Wire shapes for Chat.
@@ -206,18 +239,29 @@ const (
 // the wire shape: WireChat (OpenAI) or WireMessages (Anthropic). A non-empty
 // sessionID is sent as the universal x-session-id header the proxy records.
 func (cl *Client) Chat(ctx context.Context, endpoint, proxyIP, kind, model, prompt, sessionID string) (int, string, error) {
+	return cl.ChatPrefixed(ctx, endpoint, proxyIP, "", kind, model, prompt, sessionID)
+}
+
+// ChatPrefixed is Chat with a base-URL path prefix prepended to the wire
+// path, mirroring agents whose base URL carries a shape-selecting prefix that
+// rides through to the upstream — e.g. Claude Code against a Kimi provider
+// sets ANTHROPIC_BASE_URL=https://<endpoint>/anthropic so the proxy forwards
+// /anthropic/v1/messages to Moonshot's Anthropic surface while the provider's
+// upstream URL stays the bare https://api.moonshot.ai. Empty prefix is plain
+// Chat.
+func (cl *Client) ChatPrefixed(ctx context.Context, endpoint, proxyIP, pathPrefix, kind, model, prompt, sessionID string) (int, string, error) {
 	var path, body string
 	var headers []string
 	switch kind {
 	case WireMessages:
 		path = "/v1/messages"
 		headers = []string{"anthropic-version: 2023-06-01"}
-		body = fmt.Sprintf(`{"model":%q,"max_tokens":64,"messages":[{"role":"user","content":%q}]}`, model, prompt)
+		body = fmt.Sprintf(`{"model":%q,"max_tokens":2048,"messages":[{"role":"user","content":%q}]}`, model, prompt)
 	default:
 		path = "/v1/chat/completions"
 		body = fmt.Sprintf(`{"model":%q,"messages":[{"role":"user","content":%q}]}`, model, prompt)
 	}
-	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(headers, sessionID))
+	return cl.post(ctx, endpoint, proxyIP, pathPrefix+path, body, withSessionID(headers, sessionID))
 }
 
 // Vertex issues an Anthropic-on-Vertex rawPredict POST over the tunnel. Unlike
@@ -227,7 +271,7 @@ func (cl *Client) Chat(ctx context.Context, endpoint, proxyIP, kind, model, prom
 // is sent as the universal x-session-id header the proxy records.
 func (cl *Client) Vertex(ctx context.Context, endpoint, proxyIP, project, region, model, prompt, sessionID string) (int, string, error) {
 	path := fmt.Sprintf("/v1/projects/%s/locations/%s/publishers/anthropic/models/%s:rawPredict", project, region, model)
-	body := fmt.Sprintf(`{"anthropic_version":"vertex-2023-10-16","max_tokens":64,"messages":[{"role":"user","content":%q}]}`, prompt)
+	body := fmt.Sprintf(`{"anthropic_version":"vertex-2023-10-16","max_tokens":2048,"messages":[{"role":"user","content":%q}]}`, prompt)
 	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(nil, sessionID))
 }
 
@@ -238,7 +282,7 @@ func (cl *Client) Vertex(ctx context.Context, endpoint, proxyIP, project, region
 // header the proxy records.
 func (cl *Client) Bedrock(ctx context.Context, endpoint, proxyIP, model, prompt, sessionID string) (int, string, error) {
 	path := "/model/" + model + "/invoke"
-	body := fmt.Sprintf(`{"anthropic_version":"bedrock-2023-05-31","max_tokens":64,"messages":[{"role":"user","content":%q}]}`, prompt)
+	body := fmt.Sprintf(`{"anthropic_version":"bedrock-2023-05-31","max_tokens":2048,"messages":[{"role":"user","content":%q}]}`, prompt)
 	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(nil, sessionID))
 }
 
@@ -297,7 +341,7 @@ func (cl *Client) Terminate(ctx context.Context) error {
 	return cl.container.Terminate(ctx)
 }
 
-// containerLogs reads up to 256 KiB of a container's logs for diagnostics.
+// containerLogs reads up to 4 MiB of a container's logs for diagnostics — enough for a whole provider-matrix run.
 func containerLogs(ctx context.Context, c testcontainers.Container) string {
 	if c == nil {
 		return ""
@@ -307,6 +351,6 @@ func containerLogs(ctx context.Context, c testcontainers.Container) string {
 		return fmt.Sprintf("<logs error: %v>", err)
 	}
 	defer r.Close()
-	b, _ := io.ReadAll(io.LimitReader(r, 256<<10))
+	b, _ := io.ReadAll(io.LimitReader(r, 4<<20))
 	return string(b)
 }

e2e/harness/combined.go

@@ -221,6 +221,29 @@ func (c *Combined) CreateProxyTokenCLI(ctx context.Context, name string) (string
 	return "", fmt.Errorf("token not found in CLI output: %s", string(out))
 }
 
+// SnapshotStoreDB copies the management sqlite store (with WAL/SHM sidecars) out of the bind-mounted
+// data dir into dstDir and returns the copy's path; reading a copy avoids locking against live writes.
+func (c *Combined) SnapshotStoreDB(dstDir string) (string, error) {
+	src := filepath.Join(c.workDir, "data", "store.db")
+	if _, err := os.Stat(src); err != nil {
+		return "", fmt.Errorf("management store not found at %s: %w", src, err)
+	}
+	dst := filepath.Join(dstDir, "store.db")
+	for _, suffix := range []string{"", "-wal", "-shm"} {
+		data, err := os.ReadFile(src + suffix)
+		if err != nil {
+			if os.IsNotExist(err) && suffix != "" {
+				continue // sidecar only exists in WAL mode
+			}
+			return "", fmt.Errorf("read %s: %w", src+suffix, err)
+		}
+		if err := os.WriteFile(dst+suffix, data, 0o600); err != nil {
+			return "", fmt.Errorf("write %s: %w", dst+suffix, err)
+		}
+	}
+	return dst, nil
+}
+
 // Logs returns the combined server container logs, for diagnostics.
 func (c *Combined) Logs(ctx context.Context) string {
 	return containerLogs(ctx, c.container)

idp/dex/config.go

@@ -308,7 +308,7 @@ func (s *Storage) OpenStorage(logger *slog.Logger) (storage.Storage, error) {
 		if file == "" {
 			return nil, fmt.Errorf("sqlite3 storage requires 'file' config")
 		}
-		return (&sql.SQLite3{File: file}).Open(logger)
+		return newSQLite3(file).Open(logger)
 	case "postgres":
 		dsn, _ := s.Config["dsn"].(string)
 		if dsn == "" {

idp/dex/provider.go

@@ -20,7 +20,6 @@ import (
 	"github.com/dexidp/dex/server"
 	"github.com/dexidp/dex/server/signer"
 	"github.com/dexidp/dex/storage"
-	"github.com/dexidp/dex/storage/sql"
 	"github.com/go-jose/go-jose/v4"
 	"github.com/google/uuid"
 	"github.com/prometheus/client_golang/prometheus"
@@ -41,7 +40,7 @@ type Config struct {
 	GRPCAddr string
 }
 
-const localConnectorID = "local"
+const LocalConnectorID = "local"
 
 // Provider wraps a Dex server
 type Provider struct {
@@ -79,7 +78,7 @@ func NewProvider(ctx context.Context, config *Config) (*Provider, error) {
 
 	// Initialize SQLite storage
 	dbPath := filepath.Join(config.DataDir, "oidc.db")
-	sqliteConfig := &sql.SQLite3{File: dbPath}
+	sqliteConfig := newSQLite3(dbPath)
 	stor, err := sqliteConfig.Open(logger)
 	if err != nil {
 		return nil, fmt.Errorf("failed to open storage: %w", err)
@@ -495,18 +494,60 @@ func (p *Provider) Storage() storage.Storage {
 	return p.storage
 }
 
+// SetClientsMFAChain updates the MFAChain field on OAuth2 clients in Dex storage.
+// Pass a non-empty slice (e.g. []string{"default-totp"}) to enable MFA, or nil to disable it.
+func SetClientsMFAChain(ctx context.Context, st storage.Storage, clientIDs []string, mfaChain []string) error {
+	previousChains := make(map[string][]string, len(clientIDs))
+	for _, clientID := range clientIDs {
+		client, err := st.GetClient(ctx, clientID)
+		if err != nil {
+			return fmt.Errorf("failed to get client %s before MFA chain update: %w", clientID, err)
+		}
+		previousChains[clientID] = cloneMFAChain(client.MFAChain)
+	}
+
+	updatedClientIDs := make([]string, 0, len(clientIDs))
+	for _, clientID := range clientIDs {
+		if err := st.UpdateClient(ctx, clientID, func(old storage.Client) (storage.Client, error) {
+			old.MFAChain = cloneMFAChain(mfaChain)
+			return old, nil
+		}); err != nil {
+			if rollbackErr := rollbackClientsMFAChain(ctx, st, updatedClientIDs, previousChains); rollbackErr != nil {
+				return fmt.Errorf("failed to update MFA chain on client %s: %w (also failed to roll back previous MFA chains: %v)", clientID, err, rollbackErr)
+			}
+			return fmt.Errorf("failed to update MFA chain on client %s: %w", clientID, err)
+		}
+		updatedClientIDs = append(updatedClientIDs, clientID)
+	}
+	return nil
+}
+
+func rollbackClientsMFAChain(ctx context.Context, st storage.Storage, clientIDs []string, previousChains map[string][]string) error {
+	var rollbackErrs []error
+	for i := len(clientIDs) - 1; i >= 0; i-- {
+		clientID := clientIDs[i]
+		previousChain := cloneMFAChain(previousChains[clientID])
+		if err := st.UpdateClient(ctx, clientID, func(old storage.Client) (storage.Client, error) {
+			old.MFAChain = previousChain
+			return old, nil
+		}); err != nil {
+			rollbackErrs = append(rollbackErrs, fmt.Errorf("client %s: %w", clientID, err))
+		}
+	}
+	return errors.Join(rollbackErrs...)
+}
+
+func cloneMFAChain(chain []string) []string {
+	if chain == nil {
+		return nil
+	}
+	return append([]string(nil), chain...)
+}
+
 // SetClientsMFAChain updates the MFAChain field on the dashboard and CLI OAuth2 clients.
 // Pass a non-empty slice (e.g. []string{"default-totp"}) to enable MFA, or nil to disable it.
 func (p *Provider) SetClientsMFAChain(ctx context.Context, clientIDs []string, mfaChain []string) error {
-	for _, clientID := range clientIDs {
-		if err := p.storage.UpdateClient(ctx, clientID, func(old storage.Client) (storage.Client, error) {
-			old.MFAChain = mfaChain
-			return old, nil
-		}); err != nil {
-			return fmt.Errorf("failed to update MFA chain on client %s: %w", clientID, err)
-		}
-	}
-	return nil
+	return SetClientsMFAChain(ctx, p.storage, clientIDs, mfaChain)
 }
 
 // Handler returns the Dex server as an http.Handler for embedding in another server.
@@ -546,7 +587,7 @@ func (p *Provider) CreateUser(ctx context.Context, email, username, password str
 
 	// Encode the user ID in Dex's format: base64(protobuf{user_id, connector_id})
 	// This matches the format Dex uses in JWT tokens
-	encodedID := EncodeDexUserID(userID, localConnectorID)
+	encodedID := EncodeDexUserID(userID, LocalConnectorID)
 	return encodedID, nil
 }
 
@@ -625,7 +666,7 @@ func DecodeDexUserID(encodedID string) (userID, connectorID string, err error) {
 // local password connector.
 func IsLocalUserID(encodedID string) bool {
 	_, connectorID, err := DecodeDexUserID(encodedID)
-	return err == nil && connectorID == localConnectorID
+	return err == nil && connectorID == LocalConnectorID
 }
 
 // GetUser returns a user by email

idp/dex/provider_test.go

@@ -3,6 +3,8 @@ package dex
 import (
 	"context"
 	"encoding/json"
+	"errors"
+	"io"
 	"log/slog"
 	"net/http"
 	"net/http/httptest"
@@ -11,11 +13,44 @@ import (
 	"testing"
 
 	"github.com/dexidp/dex/storage"
+	"github.com/dexidp/dex/storage/memory"
 	sqllib "github.com/dexidp/dex/storage/sql"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
+type updateFailingStorage struct {
+	storage.Storage
+	failClientID string
+}
+
+func (s *updateFailingStorage) UpdateClient(ctx context.Context, id string, updater func(storage.Client) (storage.Client, error)) error {
+	if id == s.failClientID {
+		return errors.New("forced update failure")
+	}
+	return s.Storage.UpdateClient(ctx, id, updater)
+}
+
+func TestSetClientsMFAChainRollsBackUpdatedClients(t *testing.T) {
+	ctx := context.Background()
+	st := memory.New(slog.New(slog.NewTextHandler(io.Discard, nil)))
+
+	require.NoError(t, st.CreateClient(ctx, storage.Client{ID: "client-1", MFAChain: []string{"old-1"}}))
+	require.NoError(t, st.CreateClient(ctx, storage.Client{ID: "client-2", MFAChain: []string{"old-2"}}))
+
+	err := SetClientsMFAChain(ctx, &updateFailingStorage{Storage: st, failClientID: "client-2"}, []string{"client-1", "client-2"}, []string{"new"})
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "failed to update MFA chain on client client-2")
+
+	client1, err := st.GetClient(ctx, "client-1")
+	require.NoError(t, err)
+	require.Equal(t, []string{"old-1"}, client1.MFAChain)
+
+	client2, err := st.GetClient(ctx, "client-2")
+	require.NoError(t, err)
+	require.Equal(t, []string{"old-2"}, client2.MFAChain)
+}
+
 func TestUserCreationFlow(t *testing.T) {
 	ctx := context.Background()
 

idp/dex/sqlite_cgo.go

@@ -0,0 +1,15 @@
+//go:build cgo
+
+package dex
+
+import (
+	sql "github.com/dexidp/dex/storage/sql"
+)
+
+// newSQLite3 builds the dex SQLite3 config. CGO builds use the upstream
+// struct that takes a File path. Non-CGO builds get an empty stub whose
+// Open() returns the dex "SQLite not available" error — correct behaviour
+// for binaries that can't link sqlite3 (e.g. cross-compiled ARM targets).
+func newSQLite3(file string) *sql.SQLite3 {
+	return &sql.SQLite3{File: file}
+}

idp/dex/sqlite_nocgo.go

@@ -0,0 +1,15 @@
+//go:build !cgo
+
+package dex
+
+import (
+	sql "github.com/dexidp/dex/storage/sql"
+)
+
+// newSQLite3 for non-CGO builds. The dex SQLite3 stub has no fields and its
+// Open() returns an error documenting the missing CGO support — correct
+// behaviour for cross-compiled artefacts that never actually run the
+// embedded IdP. The `file` argument is ignored.
+func newSQLite3(_ string) *sql.SQLite3 {
+	return &sql.SQLite3{}
+}

infrastructure_files/getting-started-enterprise.sh

@@ -234,9 +234,6 @@ init_environment() {
 
   NETBIRD_LICENSE_KEY=$(read_secret "Enter license key (input hidden)")
 
-  GHCR_USERNAME="netbirdExtAccess1"
-  GHCR_TOKEN=$(read_secret "Enter GHCR token (input hidden)")
-
   POSTGRES_USER="netbird"
   POSTGRES_DB="netbird"
   POSTGRES_PASSWORD=$(rand_secret)
@@ -263,10 +260,6 @@ init_environment() {
   install -m 600 /dev/null config.yaml
   render_config_yaml >> config.yaml
 
-  echo "Logging in to ghcr.io ..."
-  printf '%s' "$GHCR_TOKEN" | docker login ghcr.io -u "$GHCR_USERNAME" --password-stdin
-  unset GHCR_TOKEN
-
   echo ""
   echo "Pulling images ..."
   $DOCKER_COMPOSE_COMMAND pull

infrastructure_files/getting-started.sh

@@ -556,7 +556,7 @@ start_services_and_show_instructions() {
       echo "Creating proxy access token..."
       # Use docker exec with bash to run the token command directly
       PROXY_TOKEN=$($DOCKER_COMPOSE_COMMAND exec -T netbird-server \
-        /go/bin/netbird-server token create --name "default-proxy" --config /etc/netbird/config.yaml 2>/dev/null | grep "^Token:" | awk '{print $2}')
+        /go/bin/netbird-server admin token create --name "default-proxy" --config /etc/netbird/config.yaml 2>/dev/null | grep "^Token:" | awk '{print $2}')
 
       if [[ -z "$PROXY_TOKEN" ]]; then
         echo "ERROR: Failed to create proxy token. Check netbird-server logs." > /dev/stderr

infrastructure_files/migrate-to-enterprise.sh

@@ -490,8 +490,6 @@ init_migration() {
   echo ""
   echo "Step 1: Image swap (community → Enterprise). License key required."
   NB_LICENSE_KEY=$(read_secret "  License key")
-  GHCR_USERNAME="netbirdExtAccess1"
-  GHCR_TOKEN=$(read_secret "  GHCR token (input hidden)")
 
   # Step 2 — optional
   echo ""
@@ -588,11 +586,6 @@ apply_changes() {
     fi
   } >> "$ENV_FILE"
 
-  echo ""
-  echo "Logging in to ghcr.io ..."
-  printf '%s' "$GHCR_TOKEN" | docker login ghcr.io -u "$GHCR_USERNAME" --password-stdin
-  unset GHCR_TOKEN
-
   echo ""
   echo "Pulling enterprise images ..."
   $DOCKER_COMPOSE_COMMAND pull

management/cmd/admin.go

@@ -0,0 +1,177 @@
+package cmd
+
+import (
+	"context"
+	"fmt"
+	"path"
+	"path/filepath"
+
+	"github.com/dexidp/dex/storage"
+	log "github.com/sirupsen/logrus"
+	"github.com/spf13/cobra"
+
+	"github.com/netbirdio/netbird/formatter/hook"
+	admincmd "github.com/netbirdio/netbird/management/cmd/admin"
+	tokencmd "github.com/netbirdio/netbird/management/cmd/token"
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
+	"github.com/netbirdio/netbird/management/server/activity"
+	activitystore "github.com/netbirdio/netbird/management/server/activity/store"
+	"github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/util"
+)
+
+var adminDatadir string
+
+// newAdminCommands creates the admin command tree with management-specific resource openers.
+func newAdminCommands() *cobra.Command {
+	cmd := admincmd.NewCommands(admincmd.Openers{
+		Resources: withAdminResources,
+		Store:     withAdminStoreOnly,
+		IDP:       withAdminIDPOnly,
+	})
+	cmd.PersistentFlags().StringVar(&adminDatadir, "datadir", "", "Override the data directory from config (used for store.db and the default idp.db)")
+	return cmd
+}
+
+func newLegacyTokenCommand() *cobra.Command {
+	cmd := tokencmd.NewCommands(tokencmd.StoreOpener(withAdminStoreOnly))
+	cmd.Deprecated = "use 'admin token' instead"
+	cmd.PersistentFlags().StringVar(&nbconfig.MgmtConfigPath, "config", defaultMgmtConfig, "Netbird config file location")
+	return cmd
+}
+
+// withAdminResources initializes logging, loads config, opens the management store
+// and embedded IdP storage, and calls fn.
+func withAdminResources(cmd *cobra.Command, fn func(ctx context.Context, resources admincmd.Resources) error) error {
+	return withAdminConfig(cmd, true, func(ctx context.Context, config *nbconfig.Config, datadir string) error {
+		managementStore, err := openAdminStore(ctx, config, datadir)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseStore(ctx, managementStore)
+
+		idpStorage, idpStorageFile, err := admincmd.OpenIDPStorage(config)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseIDPStorage(idpStorage)
+
+		eventStore, esErr := openAdminEventStore(ctx, config, datadir)
+		if esErr != nil {
+			_, _ = fmt.Fprintf(cmd.ErrOrStderr(), "Warning: audit events will not be recorded: %v\n", esErr)
+		}
+		if eventStore != nil {
+			defer func() {
+				if err := eventStore.Close(ctx); err != nil {
+					log.Debugf("close activity event store: %v", err)
+				}
+			}()
+		}
+
+		return fn(ctx, admincmd.Resources{Store: managementStore, IDPStorage: idpStorage, IDPStorageFile: idpStorageFile, EventStore: eventStore})
+	})
+}
+
+// withAdminStoreOnly opens only the management store for admin subcommands that do not
+// need embedded IdP storage.
+func withAdminStoreOnly(cmd *cobra.Command, fn func(ctx context.Context, s store.Store) error) error {
+	return withAdminConfig(cmd, false, func(ctx context.Context, config *nbconfig.Config, datadir string) error {
+		managementStore, err := openAdminStore(ctx, config, datadir)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseStore(ctx, managementStore)
+
+		return fn(ctx, managementStore)
+	})
+}
+
+func withAdminIDPOnly(cmd *cobra.Command, fn func(ctx context.Context, idpStorage storage.Storage, storageFile string) error) error {
+	return withAdminConfig(cmd, true, func(ctx context.Context, config *nbconfig.Config, _ string) error {
+		idpStorage, idpStorageFile, err := admincmd.OpenIDPStorage(config)
+		if err != nil {
+			return err
+		}
+		defer admincmd.CloseIDPStorage(idpStorage)
+
+		return fn(ctx, idpStorage, idpStorageFile)
+	})
+}
+
+func withAdminConfig(cmd *cobra.Command, applyIDPDefaults bool, fn func(ctx context.Context, config *nbconfig.Config, datadir string) error) error {
+	if err := util.InitLog("error", "console"); err != nil {
+		return fmt.Errorf("init log: %w", err)
+	}
+
+	ctx := context.WithValue(cmd.Context(), hook.ExecutionContextKey, hook.SystemSource) //nolint:staticcheck
+
+	config, datadir, err := loadAdminMgmtConfig(ctx, applyIDPDefaults)
+	if err != nil {
+		return fmt.Errorf("load config: %w", err)
+	}
+
+	return fn(ctx, config, datadir)
+}
+
+func loadAdminMgmtConfig(ctx context.Context, applyIDPDefaults bool) (*nbconfig.Config, string, error) {
+	config := &nbconfig.Config{}
+	if _, err := util.ReadJsonWithEnvSub(nbconfig.MgmtConfigPath, config); err != nil {
+		return nil, "", err
+	}
+
+	if applyIDPDefaults {
+		if err := ApplyEmbeddedIdPConfig(ctx, config); err != nil {
+			return nil, "", err
+		}
+	}
+
+	datadir := config.Datadir
+	applyAdminDatadirOverride(config, &datadir)
+	return config, datadir, nil
+}
+
+func applyAdminDatadirOverride(config *nbconfig.Config, datadir *string) {
+	if adminDatadir == "" {
+		return
+	}
+
+	oldDatadir := *datadir
+	*datadir = adminDatadir
+	if config.EmbeddedIdP != nil && config.EmbeddedIdP.Storage.Type == "sqlite3" && isDefaultIDPStorageFile(config.EmbeddedIdP.Storage.Config.File, oldDatadir) {
+		config.EmbeddedIdP.Storage.Config.File = filepath.Join(*datadir, "idp.db")
+	}
+}
+
+func isDefaultIDPStorageFile(file, datadir string) bool {
+	if file == "" {
+		return true
+	}
+	defaultFile := filepath.Join(datadir, "idp.db")
+	legacyDefaultFile := path.Join(datadir, "idp.db")
+	legacySlashDefaultFile := path.Join(filepath.ToSlash(datadir), "idp.db")
+	return filepath.Clean(file) == filepath.Clean(defaultFile) ||
+		file == legacyDefaultFile ||
+		filepath.ToSlash(file) == legacySlashDefaultFile
+}
+
+func openAdminStore(ctx context.Context, config *nbconfig.Config, datadir string) (store.Store, error) {
+	managementStore, err := store.NewStore(ctx, config.StoreConfig.Engine, datadir, nil, true)
+	if err != nil {
+		return nil, fmt.Errorf("create store: %w", err)
+	}
+	return managementStore, nil
+}
+
+func openAdminEventStore(ctx context.Context, config *nbconfig.Config, datadir string) (activity.Store, error) {
+	if config.DataStoreEncryptionKey == "" {
+		return nil, fmt.Errorf("data store encryption key is not configured")
+	}
+	eventStore, err := activitystore.NewSqlStore(ctx, datadir, config.DataStoreEncryptionKey)
+	if err != nil {
+		return nil, fmt.Errorf("open activity event store: %w", err)
+	}
+	if eventStore == nil {
+		return nil, fmt.Errorf("open activity event store: returned nil store")
+	}
+	return eventStore, nil
+}

management/cmd/admin/admin.go

@@ -0,0 +1,577 @@
+// Package admincmd provides reusable cobra commands for self-hosted administrator helpers.
+// Both the management and combined binaries use these commands, each providing
+// their own opener to handle config loading and storage initialization.
+package admincmd
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"log/slog"
+	"os"
+	"strings"
+	"time"
+
+	"github.com/dexidp/dex/storage"
+	log "github.com/sirupsen/logrus"
+	"github.com/spf13/cobra"
+	"golang.org/x/crypto/bcrypt"
+
+	"github.com/netbirdio/netbird/formatter/hook"
+	nbdex "github.com/netbirdio/netbird/idp/dex"
+	"github.com/netbirdio/netbird/management/cmd/proxy"
+	"github.com/netbirdio/netbird/management/cmd/token"
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
+	"github.com/netbirdio/netbird/management/server"
+	"github.com/netbirdio/netbird/management/server/activity"
+	"github.com/netbirdio/netbird/management/server/idp"
+	"github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/management/server/types"
+)
+
+// Resources contains the storages required by the admin commands.
+type Resources struct {
+	Store          store.Store
+	IDPStorage     storage.Storage
+	IDPStorageFile string
+	EventStore     activity.Store
+}
+
+// Opener initializes command resources from the command context and calls fn.
+type Opener func(cmd *cobra.Command, fn func(ctx context.Context, resources Resources) error) error
+
+// StoreOpener initializes only the management store from the command context and calls fn.
+type StoreOpener func(cmd *cobra.Command, fn func(ctx context.Context, s store.Store) error) error
+
+// IDPOpener initializes only the embedded IdP storage from the command context and calls fn.
+type IDPOpener func(cmd *cobra.Command, fn func(ctx context.Context, idpStorage storage.Storage, storageFile string) error) error
+
+// Openers contains the resource openers needed by the admin command tree.
+type Openers struct {
+	Resources Opener
+	Store     StoreOpener
+	IDP       IDPOpener
+}
+
+type userSelector struct {
+	email  string
+	userID string
+}
+
+func (s userSelector) normalized() userSelector {
+	return userSelector{
+		email:  strings.TrimSpace(s.email),
+		userID: strings.TrimSpace(s.userID),
+	}
+}
+
+func (s userSelector) validate() error {
+	s = s.normalized()
+	if (s.email == "") == (s.userID == "") {
+		return fmt.Errorf("provide exactly one of --email or --user-id")
+	}
+	return nil
+}
+
+// NewCommands creates the admin command tree with the given resource openers.
+func NewCommands(openers Openers) *cobra.Command {
+	adminCmd := &cobra.Command{
+		Use:   "admin",
+		Short: "Self-hosted administrator helpers",
+		Long:  "Administrative helpers for self-hosted deployments using the embedded identity provider.",
+	}
+
+	userCmd := &cobra.Command{
+		Use:   "user",
+		Short: "Manage local embedded IdP users",
+	}
+
+	var passwordSelector userSelector
+	var password string
+	var passwordFile string
+	passwordCmd := &cobra.Command{
+		Use:     "change-password (--email email | --user-id id) (--password password | --password-file path)",
+		Aliases: []string{"set-password"},
+		Short:   "Change a local user's password",
+		Args:    cobra.NoArgs,
+		RunE: func(cmd *cobra.Command, _ []string) error {
+			if err := passwordSelector.validate(); err != nil {
+				return err
+			}
+			newPassword, err := resolvePasswordInput(cmd, password, passwordFile)
+			if err != nil {
+				return err
+			}
+			return openers.IDP(cmd, func(ctx context.Context, idpStorage storage.Storage, storageFile string) error {
+				return runChangePassword(ctx, idpStorage, cmd.OutOrStdout(), passwordSelector, newPassword, storageFile)
+			})
+		},
+	}
+	addUserSelectorFlags(passwordCmd, &passwordSelector)
+	passwordCmd.Flags().StringVar(&password, "password", "", "New password for the user")
+	passwordCmd.Flags().StringVar(&passwordFile, "password-file", "", "Read new password from file ('-' for stdin)")
+
+	var resetSelector userSelector
+	resetMFACmd := &cobra.Command{
+		Use:   "reset-mfa (--email email | --user-id id)",
+		Short: "Reset a local user's MFA enrollment",
+		Args:  cobra.NoArgs,
+		RunE: func(cmd *cobra.Command, _ []string) error {
+			if err := resetSelector.validate(); err != nil {
+				return err
+			}
+			return openers.IDP(cmd, func(ctx context.Context, idpStorage storage.Storage, storageFile string) error {
+				return runResetMFA(ctx, idpStorage, cmd.OutOrStdout(), resetSelector, storageFile)
+			})
+		},
+	}
+	addUserSelectorFlags(resetMFACmd, &resetSelector)
+
+	userCmd.AddCommand(passwordCmd, resetMFACmd)
+
+	mfaCmd := &cobra.Command{
+		Use:   "mfa",
+		Short: "Manage local MFA for embedded IdP users",
+	}
+
+	enableCmd := &cobra.Command{
+		Use:   "enable",
+		Short: "Enable MFA for local embedded IdP users",
+		Args:  cobra.NoArgs,
+		RunE: func(cmd *cobra.Command, _ []string) error {
+			return openers.Resources(cmd, func(ctx context.Context, resources Resources) error {
+				return runSetMFAEnabled(ctx, resources, cmd.OutOrStdout(), true)
+			})
+		},
+	}
+
+	disableCmd := &cobra.Command{
+		Use:   "disable",
+		Short: "Disable MFA for local embedded IdP users",
+		Args:  cobra.NoArgs,
+		RunE: func(cmd *cobra.Command, _ []string) error {
+			return openers.Resources(cmd, func(ctx context.Context, resources Resources) error {
+				return runSetMFAEnabled(ctx, resources, cmd.OutOrStdout(), false)
+			})
+		},
+	}
+
+	statusCmd := &cobra.Command{
+		Use:   "status",
+		Short: "Show local MFA status",
+		Args:  cobra.NoArgs,
+		RunE: func(cmd *cobra.Command, _ []string) error {
+			return openers.Resources(cmd, func(ctx context.Context, resources Resources) error {
+				return runMFAStatus(ctx, resources, cmd.OutOrStdout())
+			})
+		},
+	}
+
+	mfaCmd.AddCommand(enableCmd, disableCmd, statusCmd)
+	adminCmd.AddCommand(userCmd, mfaCmd)
+	if openers.Store != nil {
+		adminCmd.AddCommand(tokencmd.NewCommands(tokencmd.StoreOpener(openers.Store)))
+		adminCmd.AddCommand(proxycmd.NewCommands(proxycmd.StoreOpener(openers.Store)))
+	}
+	return adminCmd
+}
+
+// OpenEmbeddedIDPStorage opens the Dex storage configured for the embedded IdP.
+func OpenEmbeddedIDPStorage(cfg *idp.EmbeddedIdPConfig) (storage.Storage, error) {
+	if cfg == nil || !cfg.Enabled {
+		return nil, fmt.Errorf("admin commands require the embedded IdP to be enabled")
+	}
+
+	yamlConfig, err := cfg.ToYAMLConfig()
+	if err != nil {
+		return nil, fmt.Errorf("build embedded IdP config: %w", err)
+	}
+
+	logger := slog.New(slog.NewTextHandler(io.Discard, nil))
+	st, err := yamlConfig.Storage.OpenStorage(logger)
+	if err != nil {
+		return nil, fmt.Errorf("open embedded IdP storage: %w", err)
+	}
+	return st, nil
+}
+
+// CloseStore closes the management store and logs cleanup errors at debug level.
+func CloseStore(ctx context.Context, s store.Store) {
+	if s == nil {
+		return
+	}
+	if err := s.Close(ctx); err != nil {
+		log.Debugf("close store: %v", err)
+	}
+}
+
+// OpenIDPStorage opens embedded IdP storage and returns its sqlite file path when applicable.
+func OpenIDPStorage(config *nbconfig.Config) (storage.Storage, string, error) {
+	if config == nil {
+		return nil, "", fmt.Errorf("management config is required")
+	}
+	idpStorage, err := OpenEmbeddedIDPStorage(config.EmbeddedIdP)
+	if err != nil {
+		return nil, "", err
+	}
+	return idpStorage, embeddedIDPStorageFile(config), nil
+}
+
+func embeddedIDPStorageFile(config *nbconfig.Config) string {
+	if config.EmbeddedIdP == nil || config.EmbeddedIdP.Storage.Type != "sqlite3" {
+		return ""
+	}
+	return config.EmbeddedIdP.Storage.Config.File
+}
+
+// CloseIDPStorage closes embedded IdP storage and logs cleanup errors at debug level.
+func CloseIDPStorage(s storage.Storage) {
+	if s == nil {
+		return
+	}
+	if err := s.Close(); err != nil {
+		log.Debugf("close embedded IdP storage: %v", err)
+	}
+}
+
+func addUserSelectorFlags(cmd *cobra.Command, selector *userSelector) {
+	cmd.Flags().StringVar(&selector.email, "email", "", "User email")
+	cmd.Flags().StringVar(&selector.userID, "user-id", "", "User ID")
+}
+
+func resolvePasswordInput(cmd *cobra.Command, password, passwordFile string) (string, error) {
+	if password != "" && passwordFile != "" {
+		return "", fmt.Errorf("provide only one of --password or --password-file")
+	}
+	if passwordFile == "" {
+		return password, nil
+	}
+
+	var data []byte
+	var err error
+	if passwordFile == "-" {
+		data, err = io.ReadAll(cmd.InOrStdin())
+	} else {
+		data, err = os.ReadFile(passwordFile)
+	}
+	if err != nil {
+		return "", fmt.Errorf("read password: %w", err)
+	}
+	return strings.TrimRight(string(data), "\r\n"), nil
+}
+
+func runChangePassword(ctx context.Context, idpStorage storage.Storage, w io.Writer, selector userSelector, password string, idpStorageFile string) error {
+	if idpStorage == nil {
+		return fmt.Errorf("embedded IdP storage is required")
+	}
+	selector = selector.normalized()
+	if err := selector.validate(); err != nil {
+		return err
+	}
+	if password == "" {
+		return fmt.Errorf("password is required")
+	}
+	if err := server.ValidatePassword(password); err != nil {
+		return fmt.Errorf("invalid password: %w", err)
+	}
+
+	user, err := findLocalUser(ctx, idpStorage, selector, idpStorageFile)
+	if err != nil {
+		return err
+	}
+
+	hash, err := bcrypt.GenerateFromPassword([]byte(password), bcrypt.DefaultCost)
+	if err != nil {
+		return fmt.Errorf("hash password: %w", err)
+	}
+
+	if err := idpStorage.UpdatePassword(ctx, user.Email, func(old storage.Password) (storage.Password, error) {
+		old.Hash = hash
+		return old, nil
+	}); err != nil {
+		return fmt.Errorf("update password for %s: %w", user.Email, err)
+	}
+
+	if err := deleteLocalAuthSession(ctx, idpStorage, user.UserID); err != nil {
+		return err
+	}
+
+	_, _ = fmt.Fprintf(w, "Password updated for %s.\n", user.Email)
+	return nil
+}
+
+func runResetMFA(ctx context.Context, idpStorage storage.Storage, w io.Writer, selector userSelector, idpStorageFile string) error {
+	if idpStorage == nil {
+		return fmt.Errorf("embedded IdP storage is required")
+	}
+	selector = selector.normalized()
+	if err := selector.validate(); err != nil {
+		return err
+	}
+
+	user, err := findLocalUser(ctx, idpStorage, selector, idpStorageFile)
+	if err != nil {
+		return err
+	}
+
+	reset := false
+	err = idpStorage.UpdateUserIdentity(ctx, user.UserID, idp.LocalConnectorID, func(old storage.UserIdentity) (storage.UserIdentity, error) {
+		reset = reset || len(old.MFASecrets) > 0 || len(old.WebAuthnCredentials) > 0
+		old.MFASecrets = map[string]*storage.MFASecret{}
+		old.WebAuthnCredentials = map[string][]storage.WebAuthnCredential{}
+		return old, nil
+	})
+	if errors.Is(err, storage.ErrNotFound) {
+		if err := deleteLocalAuthSession(ctx, idpStorage, user.UserID); err != nil {
+			return err
+		}
+		_, _ = fmt.Fprintf(w, "No MFA enrollment found for %s.\n", user.Email)
+		return nil
+	}
+	if err != nil {
+		return fmt.Errorf("reset MFA for %s: %w", user.Email, err)
+	}
+
+	if err := deleteLocalAuthSession(ctx, idpStorage, user.UserID); err != nil {
+		return err
+	}
+
+	if reset {
+		_, _ = fmt.Fprintf(w, "MFA reset for %s. The user will re-enroll at next login.\n", user.Email)
+	} else {
+		_, _ = fmt.Fprintf(w, "No MFA enrollment found for %s.\n", user.Email)
+	}
+	return nil
+}
+
+func runSetMFAEnabled(ctx context.Context, resources Resources, w io.Writer, enabled bool) error {
+	if resources.Store == nil {
+		return fmt.Errorf("management store is required")
+	}
+	if resources.IDPStorage == nil {
+		return fmt.Errorf("embedded IdP storage is required")
+	}
+
+	accountID, settings, err := getSingleAccountSettings(ctx, resources.Store)
+	if err != nil {
+		return err
+	}
+
+	oldEnabled := settings.LocalMfaEnabled
+	newSettings := settings.Copy()
+	newSettings.LocalMfaEnabled = enabled
+
+	if err := setIDPClientsMFA(ctx, resources.IDPStorage, enabled); err != nil {
+		return err
+	}
+
+	if err := resources.Store.SaveAccountSettings(ctx, accountID, newSettings); err != nil {
+		if rollbackErr := setIDPClientsMFA(ctx, resources.IDPStorage, oldEnabled); rollbackErr != nil {
+			return fmt.Errorf("save local MFA account setting: %w (also failed to roll back embedded IdP MFA state: %v)", err, rollbackErr)
+		}
+		return fmt.Errorf("save local MFA account setting: %w", err)
+	}
+
+	if err := storeMFAActivity(ctx, resources.EventStore, accountID, enabled); err != nil {
+		_, _ = fmt.Fprintf(w, "Warning: failed to record audit event: %v\n", err)
+	}
+
+	state := "disabled"
+	if enabled {
+		state = "enabled"
+	}
+	_, _ = fmt.Fprintf(w, "Local MFA %s.\n", state)
+	return nil
+}
+
+func runMFAStatus(ctx context.Context, resources Resources, w io.Writer) error {
+	if resources.Store == nil {
+		return fmt.Errorf("management store is required")
+	}
+	if resources.IDPStorage == nil {
+		return fmt.Errorf("embedded IdP storage is required")
+	}
+
+	_, settings, err := getSingleAccountSettings(ctx, resources.Store)
+	if err != nil {
+		return err
+	}
+	accountStatus := "disabled"
+	if settings.LocalMfaEnabled {
+		accountStatus = "enabled"
+	}
+
+	clientStatus, err := idpClientsMFAStatus(ctx, resources.IDPStorage)
+	if err != nil {
+		return err
+	}
+
+	_, _ = fmt.Fprintf(w, "Account setting: %s\n", accountStatus)
+	_, _ = fmt.Fprintf(w, "Embedded IdP clients: %s\n", clientStatus)
+	return nil
+}
+
+func getSingleAccountSettings(ctx context.Context, s store.Store) (string, *types.Settings, error) {
+	count, err := s.GetAccountsCounter(ctx)
+	if err != nil {
+		return "", nil, fmt.Errorf("count accounts: %w", err)
+	}
+	if count != 1 {
+		return "", nil, fmt.Errorf("expected exactly one account, got %d; local MFA is supported only in single-account embedded IdP deployments", count)
+	}
+
+	accountID, err := s.GetAnyAccountID(ctx)
+	if err != nil {
+		return "", nil, fmt.Errorf("get account ID: %w", err)
+	}
+
+	settings, err := s.GetAccountSettings(ctx, store.LockingStrengthNone, accountID)
+	if err != nil {
+		return "", nil, fmt.Errorf("get account settings: %w", err)
+	}
+	if settings == nil {
+		settings = &types.Settings{}
+	}
+	return accountID, settings, nil
+}
+
+func storeMFAActivity(ctx context.Context, eventStore activity.Store, accountID string, enabled bool) error {
+	if eventStore == nil {
+		return nil
+	}
+	event := activity.AccountLocalMfaDisabled
+	if enabled {
+		event = activity.AccountLocalMfaEnabled
+	}
+	_, err := eventStore.Save(ctx, &activity.Event{
+		Timestamp:   time.Now().UTC(),
+		Activity:    event,
+		InitiatorID: string(hook.SystemSource),
+		TargetID:    accountID,
+		AccountID:   accountID,
+	})
+	if err != nil {
+		return fmt.Errorf("save local MFA audit event: %w", err)
+	}
+	return nil
+}
+
+func findLocalUser(ctx context.Context, idpStorage storage.Storage, selector userSelector, idpStorageFile string) (storage.Password, error) {
+	selector = selector.normalized()
+	if err := selector.validate(); err != nil {
+		return storage.Password{}, err
+	}
+
+	if selector.email != "" {
+		user, err := idpStorage.GetPassword(ctx, selector.email)
+		if errors.Is(err, storage.ErrNotFound) {
+			if empty, listErr := localUsersEmpty(ctx, idpStorage); listErr != nil {
+				return storage.Password{}, listErr
+			} else if empty {
+				return storage.Password{}, noLocalUsersError(idpStorageFile)
+			}
+			return storage.Password{}, fmt.Errorf("local user with email %q not found", selector.email)
+		}
+		if err != nil {
+			return storage.Password{}, fmt.Errorf("get local user by email %q: %w", selector.email, err)
+		}
+		return user, nil
+	}
+
+	rawUserID := selector.userID
+	if decodedUserID, _, err := nbdex.DecodeDexUserID(selector.userID); err == nil && decodedUserID != "" {
+		rawUserID = decodedUserID
+	}
+
+	users, err := idpStorage.ListPasswords(ctx)
+	if err != nil {
+		return storage.Password{}, fmt.Errorf("list local users: %w", err)
+	}
+	for _, user := range users {
+		if user.UserID == rawUserID || user.UserID == selector.userID {
+			return user, nil
+		}
+	}
+
+	if len(users) == 0 {
+		return storage.Password{}, noLocalUsersError(idpStorageFile)
+	}
+
+	return storage.Password{}, fmt.Errorf("local user with ID %q not found", selector.userID)
+}
+
+func localUsersEmpty(ctx context.Context, idpStorage storage.Storage) (bool, error) {
+	users, err := idpStorage.ListPasswords(ctx)
+	if err != nil {
+		return false, fmt.Errorf("list local users: %w", err)
+	}
+	return len(users) == 0, nil
+}
+
+func noLocalUsersError(idpStorageFile string) error {
+	location := ""
+	if idpStorageFile != "" {
+		location = fmt.Sprintf(" (%s)", idpStorageFile)
+	}
+	return fmt.Errorf("no local users exist in the embedded IdP storage%s; the management server may never have started with this config, or --datadir points at the wrong location", location)
+}
+
+func deleteLocalAuthSession(ctx context.Context, idpStorage storage.Storage, userID string) error {
+	err := idpStorage.DeleteAuthSession(ctx, userID, idp.LocalConnectorID)
+	if err == nil || errors.Is(err, storage.ErrNotFound) {
+		return nil
+	}
+	return fmt.Errorf("delete local auth session for user %s: %w", userID, err)
+}
+
+func setIDPClientsMFA(ctx context.Context, idpStorage storage.Storage, enabled bool) error {
+	var mfaChain []string
+	if enabled {
+		mfaChain = []string{idp.DefaultTOTPAuthenticatorID}
+	}
+
+	clientIDs := []string{idp.StaticClientCLI, idp.StaticClientDashboard}
+	if err := nbdex.SetClientsMFAChain(ctx, idpStorage, clientIDs, mfaChain); err != nil {
+		if errors.Is(err, storage.ErrNotFound) {
+			return fmt.Errorf("embedded IdP client not found; start the management server once before toggling MFA: %w", err)
+		}
+		return fmt.Errorf("update MFA chain on embedded IdP clients: %w", err)
+	}
+	return nil
+}
+
+func idpClientsMFAStatus(ctx context.Context, idpStorage storage.Storage) (string, error) {
+	clientIDs := []string{idp.StaticClientCLI, idp.StaticClientDashboard}
+	enabledCount := 0
+	for _, clientID := range clientIDs {
+		client, err := idpStorage.GetClient(ctx, clientID)
+		if errors.Is(err, storage.ErrNotFound) {
+			return "unknown", fmt.Errorf("embedded IdP client %q not found", clientID)
+		}
+		if err != nil {
+			return "unknown", fmt.Errorf("get embedded IdP client %q: %w", clientID, err)
+		}
+		if hasAuthenticator(client.MFAChain, idp.DefaultTOTPAuthenticatorID) {
+			enabledCount++
+		}
+	}
+
+	switch enabledCount {
+	case 0:
+		return "disabled", nil
+	case len(clientIDs):
+		return "enabled", nil
+	default:
+		return "partially enabled", nil
+	}
+}
+
+func hasAuthenticator(chain []string, authenticatorID string) bool {
+	for _, id := range chain {
+		if id == authenticatorID {
+			return true
+		}
+	}
+	return false
+}

management/cmd/admin/admin_test.go

@@ -0,0 +1,250 @@
+package admincmd
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"log/slog"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/dexidp/dex/storage"
+	"github.com/dexidp/dex/storage/memory"
+	"github.com/spf13/cobra"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/crypto/bcrypt"
+
+	nbdex "github.com/netbirdio/netbird/idp/dex"
+	"github.com/netbirdio/netbird/management/server/idp"
+	mgmtstore "github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/management/server/types"
+)
+
+func newTestIDPStorage(t *testing.T) storage.Storage {
+	t.Helper()
+
+	st := memory.New(slog.New(slog.NewTextHandler(io.Discard, nil)))
+	hash, err := bcrypt.GenerateFromPassword([]byte("OldPass1!"), bcrypt.DefaultCost)
+	require.NoError(t, err)
+
+	require.NoError(t, st.CreatePassword(context.Background(), storage.Password{
+		Email:    "user@example.com",
+		Username: "User",
+		UserID:   "user-1",
+		Hash:     hash,
+	}))
+	require.NoError(t, st.CreateUserIdentity(context.Background(), storage.UserIdentity{
+		UserID:      "user-1",
+		ConnectorID: idp.LocalConnectorID,
+		MFASecrets: map[string]*storage.MFASecret{
+			idp.DefaultTOTPAuthenticatorID: {
+				AuthenticatorID: idp.DefaultTOTPAuthenticatorID,
+				Type:            "TOTP",
+				Secret:          "otpauth://totp/NetBird:user@example.com?secret=ABC",
+				Confirmed:       true,
+				CreatedAt:       time.Now(),
+			},
+		},
+		WebAuthnCredentials: map[string][]storage.WebAuthnCredential{
+			"webauthn": {{CredentialID: []byte("credential")}},
+		},
+	}))
+	require.NoError(t, st.CreateAuthSession(context.Background(), storage.AuthSession{
+		UserID:      "user-1",
+		ConnectorID: idp.LocalConnectorID,
+		Nonce:       "nonce",
+	}))
+	require.NoError(t, st.CreateClient(context.Background(), storage.Client{ID: idp.StaticClientCLI, Name: "CLI"}))
+	require.NoError(t, st.CreateClient(context.Background(), storage.Client{ID: idp.StaticClientDashboard, Name: "Dashboard"}))
+
+	return st
+}
+
+func TestRunChangePassword(t *testing.T) {
+	ctx := context.Background()
+	st := newTestIDPStorage(t)
+	var out bytes.Buffer
+
+	err := runChangePassword(ctx, st, &out, userSelector{email: "user@example.com"}, "NewPass1!", "")
+	require.NoError(t, err)
+	require.Contains(t, out.String(), "Password updated")
+
+	user, err := st.GetPassword(ctx, "user@example.com")
+	require.NoError(t, err)
+	require.NoError(t, bcrypt.CompareHashAndPassword(user.Hash, []byte("NewPass1!")))
+
+	_, err = st.GetAuthSession(ctx, "user-1", idp.LocalConnectorID)
+	require.ErrorIs(t, err, storage.ErrNotFound)
+}
+
+func TestRunChangePasswordValidatesPassword(t *testing.T) {
+	st := newTestIDPStorage(t)
+	err := runChangePassword(context.Background(), st, io.Discard, userSelector{email: "user@example.com"}, "short", "")
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "invalid password")
+}
+
+func TestRunResetMFA(t *testing.T) {
+	ctx := context.Background()
+	st := newTestIDPStorage(t)
+	var out bytes.Buffer
+
+	encodedUserID := nbdex.EncodeDexUserID("user-1", idp.LocalConnectorID)
+	err := runResetMFA(ctx, st, &out, userSelector{userID: encodedUserID}, "")
+	require.NoError(t, err)
+	require.Contains(t, out.String(), "MFA reset")
+
+	identity, err := st.GetUserIdentity(ctx, "user-1", idp.LocalConnectorID)
+	require.NoError(t, err)
+	require.Empty(t, identity.MFASecrets)
+	require.Empty(t, identity.WebAuthnCredentials)
+
+	_, err = st.GetAuthSession(ctx, "user-1", idp.LocalConnectorID)
+	require.ErrorIs(t, err, storage.ErrNotFound)
+}
+
+func TestRunResetMFAWithoutEnrollment(t *testing.T) {
+	ctx := context.Background()
+	st := newTestIDPStorage(t)
+	require.NoError(t, st.UpdateUserIdentity(ctx, "user-1", idp.LocalConnectorID, func(old storage.UserIdentity) (storage.UserIdentity, error) {
+		old.MFASecrets = nil
+		old.WebAuthnCredentials = nil
+		return old, nil
+	}))
+
+	var out bytes.Buffer
+	err := runResetMFA(ctx, st, &out, userSelector{email: "user@example.com"}, "")
+	require.NoError(t, err)
+	require.Contains(t, out.String(), "No MFA enrollment found")
+}
+
+func TestSetIDPClientsMFA(t *testing.T) {
+	ctx := context.Background()
+	st := newTestIDPStorage(t)
+
+	require.NoError(t, setIDPClientsMFA(ctx, st, true))
+	status, err := idpClientsMFAStatus(ctx, st)
+	require.NoError(t, err)
+	require.Equal(t, "enabled", status)
+
+	require.NoError(t, setIDPClientsMFA(ctx, st, false))
+	status, err = idpClientsMFAStatus(ctx, st)
+	require.NoError(t, err)
+	require.Equal(t, "disabled", status)
+}
+
+func newTestManagementStore(t *testing.T, localMFAEnabled bool) mgmtstore.Store {
+	t.Helper()
+	ctx := context.Background()
+	st, err := mgmtstore.NewStore(ctx, types.SqliteStoreEngine, t.TempDir(), nil, false)
+	require.NoError(t, err)
+	t.Cleanup(func() { require.NoError(t, st.Close(ctx)) })
+	require.NoError(t, st.SaveAccount(ctx, &types.Account{
+		Id:       "account-1",
+		Settings: &types.Settings{LocalMfaEnabled: localMFAEnabled},
+	}))
+	return st
+}
+
+func TestRunSetMFAEnabledDoesNotSaveWhenIDPUpdateFails(t *testing.T) {
+	ctx := context.Background()
+	managementStore := newTestManagementStore(t, false)
+	idpStorage := memory.New(slog.New(slog.NewTextHandler(io.Discard, nil)))
+
+	err := runSetMFAEnabled(ctx, Resources{Store: managementStore, IDPStorage: idpStorage}, io.Discard, true)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "embedded IdP client")
+
+	settings, err := managementStore.GetAccountSettings(ctx, mgmtstore.LockingStrengthNone, "account-1")
+	require.NoError(t, err)
+	require.False(t, settings.LocalMfaEnabled)
+}
+
+func TestRunSetMFAEnabledUpdatesSettingsAfterIDP(t *testing.T) {
+	ctx := context.Background()
+	managementStore := newTestManagementStore(t, false)
+	idpStorage := newTestIDPStorage(t)
+
+	err := runSetMFAEnabled(ctx, Resources{Store: managementStore, IDPStorage: idpStorage}, io.Discard, true)
+	require.NoError(t, err)
+
+	settings, err := managementStore.GetAccountSettings(ctx, mgmtstore.LockingStrengthNone, "account-1")
+	require.NoError(t, err)
+	require.True(t, settings.LocalMfaEnabled)
+	clientStatus, err := idpClientsMFAStatus(ctx, idpStorage)
+	require.NoError(t, err)
+	require.Equal(t, "enabled", clientStatus)
+}
+
+func TestRunSetMFAEnabledSucceedsWithNilEventStore(t *testing.T) {
+	ctx := context.Background()
+	managementStore := newTestManagementStore(t, false)
+	idpStorage := newTestIDPStorage(t)
+	var out bytes.Buffer
+	var err error
+
+	require.NotPanics(t, func() {
+		err = runSetMFAEnabled(ctx, Resources{Store: managementStore, IDPStorage: idpStorage, EventStore: nil}, &out, true)
+	})
+	require.NoError(t, err)
+	require.Contains(t, out.String(), "Local MFA enabled")
+
+	settings, err := managementStore.GetAccountSettings(ctx, mgmtstore.LockingStrengthNone, "account-1")
+	require.NoError(t, err)
+	require.True(t, settings.LocalMfaEnabled)
+}
+
+func TestUserSelectorValidate(t *testing.T) {
+	require.NoError(t, userSelector{email: " user@example.com "}.validate())
+	require.NoError(t, userSelector{userID: "user-1"}.validate())
+	require.Error(t, userSelector{}.validate())
+	require.Error(t, userSelector{email: "user@example.com", userID: "user-1"}.validate())
+}
+
+func TestFindLocalUserNotFound(t *testing.T) {
+	st := newTestIDPStorage(t)
+	_, err := findLocalUser(context.Background(), st, userSelector{email: "missing@example.com"}, "")
+	require.Error(t, err)
+	require.True(t, strings.Contains(err.Error(), "not found"))
+}
+
+func TestFindLocalUserZeroUsersIncludesStoragePath(t *testing.T) {
+	st := memory.New(slog.New(slog.NewTextHandler(io.Discard, nil)))
+	_, err := findLocalUser(context.Background(), st, userSelector{email: "missing@example.com"}, "/var/lib/netbird/idp.db")
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "no local users exist")
+	require.Contains(t, err.Error(), "/var/lib/netbird/idp.db")
+}
+
+func TestUserCommandValidatesSelectorBeforeOpeningStorage(t *testing.T) {
+	opened := false
+	cmd := NewCommands(Openers{
+		IDP: func(cmd *cobra.Command, fn func(ctx context.Context, idpStorage storage.Storage, storageFile string) error) error {
+			opened = true
+			return nil
+		},
+	})
+	cmd.SetArgs([]string{"user", "change-password", "--password", "NewPass1!"})
+	cmd.SetOut(io.Discard)
+	cmd.SetErr(io.Discard)
+
+	err := cmd.Execute()
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "provide exactly one")
+	require.False(t, opened)
+}
+
+func TestResolvePasswordInputFromStdin(t *testing.T) {
+	cmd := &cobra.Command{}
+	cmd.SetIn(strings.NewReader("NewPass1!\n"))
+
+	password, err := resolvePasswordInput(cmd, "", "-")
+	require.NoError(t, err)
+	require.Equal(t, "NewPass1!", password)
+}
+
+func TestResolvePasswordInputRejectsMultipleSources(t *testing.T) {
+	_, err := resolvePasswordInput(&cobra.Command{}, "NewPass1!", "-")
+	require.Error(t, err)
+}

management/cmd/admin_config_test.go

@@ -0,0 +1,80 @@
+package cmd
+
+import (
+	"context"
+	"path"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
+	"github.com/netbirdio/netbird/management/server/idp"
+)
+
+func TestApplyAdminDatadirOverrideRelocatesDefaultIDPStorage(t *testing.T) {
+	oldDatadir := filepath.Join(t.TempDir(), "old")
+	newDatadir := filepath.Join(t.TempDir(), "new")
+
+	for _, defaultFile := range []string{
+		"",
+		filepath.Join(oldDatadir, "idp.db"),
+		path.Join(oldDatadir, "idp.db"),
+	} {
+		t.Run(defaultFile, func(t *testing.T) {
+			cfg := &nbconfig.Config{
+				EmbeddedIdP: &idp.EmbeddedIdPConfig{
+					Enabled: true,
+					Storage: idp.EmbeddedStorageConfig{
+						Type: "sqlite3",
+						Config: idp.EmbeddedStorageTypeConfig{
+							File: defaultFile,
+						},
+					},
+				},
+			}
+			datadir := oldDatadir
+			oldAdminDatadir := adminDatadir
+			adminDatadir = newDatadir
+			t.Cleanup(func() { adminDatadir = oldAdminDatadir })
+
+			applyAdminDatadirOverride(cfg, &datadir)
+
+			require.Equal(t, newDatadir, datadir)
+			require.Equal(t, filepath.Join(newDatadir, "idp.db"), cfg.EmbeddedIdP.Storage.Config.File)
+		})
+	}
+}
+
+func TestOpenAdminEventStoreMissingEncryptionKeyReturnsNilInterface(t *testing.T) {
+	eventStore, err := openAdminEventStore(context.Background(), &nbconfig.Config{}, t.TempDir())
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "encryption key")
+	require.Nil(t, eventStore)
+}
+
+func TestApplyAdminDatadirOverrideKeepsExplicitIDPStorage(t *testing.T) {
+	oldDatadir := filepath.Join(t.TempDir(), "old")
+	newDatadir := filepath.Join(t.TempDir(), "new")
+	explicitFile := filepath.Join(t.TempDir(), "custom-idp.db")
+	cfg := &nbconfig.Config{
+		EmbeddedIdP: &idp.EmbeddedIdPConfig{
+			Enabled: true,
+			Storage: idp.EmbeddedStorageConfig{
+				Type: "sqlite3",
+				Config: idp.EmbeddedStorageTypeConfig{
+					File: explicitFile,
+				},
+			},
+		},
+	}
+	datadir := oldDatadir
+	oldAdminDatadir := adminDatadir
+	adminDatadir = newDatadir
+	t.Cleanup(func() { adminDatadir = oldAdminDatadir })
+
+	applyAdminDatadirOverride(cfg, &datadir)
+
+	require.Equal(t, newDatadir, datadir)
+	require.Equal(t, explicitFile, cfg.EmbeddedIdP.Storage.Config.File)
+}

management/cmd/management.go

@@ -13,6 +13,7 @@ import (
 	"os"
 	"os/signal"
 	"path"
+	"path/filepath"
 	"strings"
 	"syscall"
 
@@ -25,6 +26,7 @@ import (
 	"github.com/netbirdio/netbird/management/internals/server"
 	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
 	nbdomain "github.com/netbirdio/netbird/shared/management/domain"
+	"github.com/netbirdio/netbird/shared/management/grpc"
 	"github.com/netbirdio/netbird/util"
 	"github.com/netbirdio/netbird/util/crypt"
 )
@@ -153,8 +155,20 @@ func LoadMgmtConfig(ctx context.Context, mgmtConfigPath string) (*nbconfig.Confi
 
 	ApplyCommandLineOverrides(loadedConfig)
 
+	err := grpc.ValidateSyncMessageVersion(loadedConfig.HighestSupportedSyncMessageVersion)
+	if err != nil {
+		return nil, err
+	}
+
+	for account, version := range loadedConfig.PerAccountHighestSupportedSyncMessageVersion {
+		err := grpc.ValidateSyncMessageVersion(&version)
+		if err != nil {
+			return nil, fmt.Errorf("unrecognized sync message version for account %s, %w", account, err)
+		}
+	}
+
 	// Apply EmbeddedIdP config to HttpConfig if embedded IdP is enabled
-	err := ApplyEmbeddedIdPConfig(ctx, loadedConfig)
+	err = ApplyEmbeddedIdPConfig(ctx, loadedConfig)
 	if err != nil {
 		return nil, err
 	}
@@ -209,7 +223,7 @@ func ApplyEmbeddedIdPConfig(ctx context.Context, cfg *nbconfig.Config) error {
 		cfg.EmbeddedIdP.Storage.Type = "sqlite3"
 	}
 	if cfg.EmbeddedIdP.Storage.Config.File == "" && cfg.Datadir != "" {
-		cfg.EmbeddedIdP.Storage.Config.File = path.Join(cfg.Datadir, "idp.db")
+		cfg.EmbeddedIdP.Storage.Config.File = filepath.Join(cfg.Datadir, "idp.db")
 	}
 
 	issuer := cfg.EmbeddedIdP.Issuer