[management,client] 0.74.5 branch sync (#6769)

## Describe your changes
* [proxy] enforce model allowlist for URL-routed providers
(Bedrock/Vertex) by @mlsmaycon in
https://github.com/netbirdio/netbird/pull/6764
* [management] Remove proxy peer stale deduplication logic by @mlsmaycon
in https://github.com/netbirdio/netbird/pull/6768
## Issue ticket number and link

## Stack

<!-- branch-stack -->

### Checklist
- [ ] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [ ] Created tests that fail without the change (if possible)
- [ ] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation
Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

### Docs PR URL (required if "docs added" is checked)
Paste the PR link from https://github.com/netbirdio/docs here:

https://github.com/netbirdio/docs/pull/__


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **New Features**
- Added model-allowlist guardrails for path-routed providers, including
Bedrock and Vertex.
  - Added Bedrock request support for chat interactions.
  - Added guardrail management capabilities.

- **Bug Fixes**
- Requests with missing or blank model identifiers are now denied when a
model allowlist is configured, improving fail-closed protection.
- Corrected provider-specific request handling and session tracking for
Bedrock interactions.

- **Tests**
- Expanded coverage for allowlist enforcement and provider routing
scenarios.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Theodor Midtlien <theodor@midtlien.com>
Co-authored-by: blaugrau90 <61945343+blaugrau90@users.noreply.github.com>
Co-authored-by: Viktor Liu <17948409+lixmal@users.noreply.github.com>

client/cmd/login.go

@@ -17,7 +17,9 @@ import (
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/auth"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	nbnet "github.com/netbirdio/netbird/client/net"
 	"github.com/netbirdio/netbird/client/proto"
+	"github.com/netbirdio/netbird/client/server"
 	"github.com/netbirdio/netbird/client/system"
 	"github.com/netbirdio/netbird/util"
 )
@@ -331,6 +333,14 @@ func doForegroundLogin(ctx context.Context, cmd *cobra.Command, setupKey string,
 		return fmt.Errorf("read config file %s: %v", configFilePath, err)
 	}
 
+	// Mirror runInForegroundMode: recover residual state (DNS, firewall,
+	// ssh config, legacy routing) from a previous unclean shutdown and
+	// enable advanced routing before dialing management.
+	if err := server.RestoreResidualState(ctx, profilemanager.NewServiceManager(configFilePath).GetStatePath()); err != nil {
+		log.Warnf("failed to restore residual state: %v", err)
+	}
+	nbnet.Init()
+
 	err = foregroundLogin(ctx, cmd, config, setupKey, activeProf.ID)
 	if err != nil {
 		return fmt.Errorf("foreground login failed: %v", err)

client/cmd/up.go

@@ -22,6 +22,8 @@ import (
 	"github.com/netbirdio/netbird/client/internal/peer"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
 	"github.com/netbirdio/netbird/client/proto"
+	nbnet "github.com/netbirdio/netbird/client/net"
+	"github.com/netbirdio/netbird/client/server"
 	"github.com/netbirdio/netbird/client/system"
 	"github.com/netbirdio/netbird/shared/management/domain"
 	"github.com/netbirdio/netbird/util"
@@ -229,6 +231,24 @@ func runInForegroundMode(ctx context.Context, cmd *cobra.Command, activeProf *pr
 
 	_, _ = profilemanager.UpdateOldManagementURL(ctx, config, configFilePath)
 
+	// Restore residual state left by a previous run that did not shut down
+	// cleanly, mirroring what the daemon does before connecting: it recovers
+	// DNS config (a stale resolv.conf takeover can make the management
+	// hostname unresolvable), firewall rules, ssh config and legacy routing.
+	// Route cleanup itself happens at engine start; nbnet.Init() below lets
+	// the management dial bypass a leftover fwmark rule until then.
+	// Foreground mode is particularly exposed in containers: a crashed
+	// container restarts inside the same (pod) network namespace, so stale
+	// state survives while the process does not.
+	if err := server.RestoreResidualState(ctx, profilemanager.NewServiceManager(configPath).GetStatePath()); err != nil {
+		log.Warnf("failed to restore residual state: %v", err)
+	}
+
+	// Enable advanced routing (as the daemon does on startup) so the
+	// management dial bypasses a leftover fwmark rule instead of being
+	// shunted into a stale routing table.
+	nbnet.Init()
+
 	err = foregroundLogin(ctx, cmd, config, providedSetupKey, activeProf.ID)
 	if err != nil {
 		return fmt.Errorf("foreground login failed: %v", err)

client/internal/debug/debug.go

@@ -480,7 +480,6 @@ func (g *BundleGenerator) addStatus() error {
 
 		fullStatus := g.statusRecorder.GetFullStatus()
 		protoFullStatus := nbstatus.ToProtoFullStatus(fullStatus)
-		protoFullStatus.Events = g.statusRecorder.GetEventHistory()
 		overview := nbstatus.ConvertToStatusOutputOverview(protoFullStatus, nbstatus.ConvertOptions{
 			Anonymize:     g.anonymize,
 			ProfileName:   profName,

client/internal/engine.go

@@ -2605,13 +2605,14 @@ func (e *Engine) updateForwardRules(rules []*mgmProto.ForwardingRule) ([]firewal
 
 func (e *Engine) toExcludedLazyPeers(rules []firewallManager.ForwardRule, peers []*mgmProto.RemotePeerConfig) map[string]bool {
 	excludedPeers := make(map[string]bool)
+
+	// Ingress forward targets: inbound forwarded traffic is initiated remotely and
+	// cannot wake a lazy connection, so the peer routing the target must stay
+	// permanently connected. AllowedIPs are already parsed on the peer conn, so
+	// reuse those typed prefixes instead of re-parsing the network map strings.
 	for _, r := range rules {
-		ip := r.TranslatedAddress
 		for _, p := range peers {
-			for _, allowedIP := range p.GetAllowedIps() {
-				if allowedIP != ip.String() {
-					continue
-				}
+			if e.peerRoutesAddr(p, r.TranslatedAddress) {
 				log.Infof("exclude forwarder peer from lazy connection: %s", p.GetWgPubKey())
 				excludedPeers[p.GetWgPubKey()] = true
 			}
@@ -2621,6 +2622,27 @@ func (e *Engine) toExcludedLazyPeers(rules []firewallManager.ForwardRule, peers
 	return excludedPeers
 }
 
+// peerRoutesAddr reports whether the peer is a router for addr, matched against
+// the peer's already-parsed AllowedIPs from the store (the same typed value the
+// lazy manager consumes) rather than re-parsing the network map strings.
+func (e *Engine) peerRoutesAddr(p *mgmProto.RemotePeerConfig, addr netip.Addr) bool {
+	prefixes, ok := e.peerStore.AllowedIPs(p.GetWgPubKey())
+	if !ok {
+		return false
+	}
+	return prefixesContain(prefixes, addr)
+}
+
+// prefixesContain reports whether addr falls within any of the prefixes.
+func prefixesContain(prefixes []netip.Prefix, addr netip.Addr) bool {
+	for _, prefix := range prefixes {
+		if prefix.Contains(addr) {
+			return true
+		}
+	}
+	return false
+}
+
 // isChecksEqual checks if two slices of checks are equal.
 func isChecksEqual(checks1, checks2 []*mgmProto.Checks) bool {
 	normalize := func(checks []*mgmProto.Checks) []string {

client/internal/engine_lazy_exclude_test.go

@@ -0,0 +1,87 @@
+package internal
+
+import (
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	firewallManager "github.com/netbirdio/netbird/client/firewall/manager"
+	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/client/internal/peerstore"
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+func TestPrefixesContain(t *testing.T) {
+	tests := []struct {
+		name     string
+		prefixes []string
+		addr     string
+		want     bool
+	}{
+		{name: "own overlay /32 matches", prefixes: []string{"100.110.8.145/32"}, addr: "100.110.8.145", want: true},
+		{name: "addr inside routed subnet", prefixes: []string{"10.121.0.0/16"}, addr: "10.121.208.4", want: true},
+		{name: "addr outside subnet", prefixes: []string{"10.121.0.0/16"}, addr: "10.122.0.1", want: false},
+		{name: "different /32", prefixes: []string{"100.110.8.145/32"}, addr: "100.110.8.146", want: false},
+		{name: "ipv6 /128 matches", prefixes: []string{"fd00::1/128"}, addr: "fd00::1", want: true},
+		{name: "no prefixes", prefixes: nil, addr: "10.121.208.4", want: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			prefixes := make([]netip.Prefix, 0, len(tt.prefixes))
+			for _, p := range tt.prefixes {
+				prefixes = append(prefixes, netip.MustParsePrefix(p))
+			}
+			require.Equal(t, tt.want, prefixesContain(prefixes, netip.MustParseAddr(tt.addr)))
+		})
+	}
+}
+
+// TestToExcludedLazyPeers_ForwardTarget guards a regression: the forward-target
+// peer (the peer routing a ForwardRule.TranslatedAddress) must be excluded from
+// lazy connections, matched via the peer's already-parsed AllowedIPs.
+func TestToExcludedLazyPeers_ForwardTarget(t *testing.T) {
+	const targetPeerKey = "cccccccccccccccccccccccccccccccccccccccccc0="
+	const otherPeerKey = "dddddddddddddddddddddddddddddddddddddddddd0="
+
+	store := peerstore.NewConnStore()
+	store.AddPeerConn(targetPeerKey, newTestConn(t, targetPeerKey, "100.110.8.145/32"))
+	store.AddPeerConn(otherPeerKey, newTestConn(t, otherPeerKey, "100.110.9.10/32"))
+
+	e := &Engine{peerStore: store}
+
+	peers := []*mgmProto.RemotePeerConfig{
+		{WgPubKey: targetPeerKey, AllowedIps: []string{"100.110.8.145/32"}},
+		{WgPubKey: otherPeerKey, AllowedIps: []string{"100.110.9.10/32"}},
+	}
+	rules := []firewallManager.ForwardRule{
+		{TranslatedAddress: netip.MustParseAddr("100.110.8.145")},
+	}
+
+	excluded := e.toExcludedLazyPeers(rules, peers)
+
+	require.True(t, excluded[targetPeerKey], "forward-target peer must be excluded from lazy connections")
+	require.False(t, excluded[otherPeerKey], "non-target peer must not be excluded")
+	require.Len(t, excluded, 1)
+}
+
+func TestToExcludedLazyPeers_NoRules(t *testing.T) {
+	e := &Engine{peerStore: peerstore.NewConnStore()}
+
+	peers := []*mgmProto.RemotePeerConfig{
+		{WgPubKey: "peer-a", AllowedIps: []string{"100.110.8.145/32"}},
+	}
+
+	require.Empty(t, e.toExcludedLazyPeers(nil, peers))
+}
+
+func newTestConn(t *testing.T, key, allowedIP string) *peer.Conn {
+	t.Helper()
+	conn, err := peer.NewConn(peer.ConnConfig{
+		Key:      key,
+		WgConfig: peer.WgConfig{AllowedIps: []netip.Prefix{netip.MustParsePrefix(allowedIP)}},
+	}, peer.ServiceDependencies{})
+	require.NoError(t, err)
+	return conn
+}

client/internal/peer/conn.go

@@ -203,7 +203,6 @@ func NewConn(config ConnConfig, services ServiceDependencies) (*Conn, error) {
 		statusICE:          worker.NewAtomicStatus(),
 		dumpState:          dumpState,
 		endpointUpdater:    NewEndpointUpdater(connLog, config.WgConfig, isController(config)),
-		wgWatcher:          NewWGWatcher(connLog, config.WgConfig.WgInterface, config.Key, dumpState),
 		metricsRecorder:    services.MetricsRecorder,
 	}
 
@@ -671,11 +670,12 @@ func (conn *Conn) onGuardEvent() {
 	}
 }
 
-func (conn *Conn) onWGDisconnected() {
+func (conn *Conn) onWGDisconnected(watcherCtx context.Context) {
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
 
-	if conn.ctx.Err() != nil {
+	// watcherCtx guards against a stale watcher tearing down a connection that already superseded it.
+	if conn.ctx.Err() != nil || watcherCtx.Err() != nil {
 		return
 	}
 
@@ -833,25 +833,39 @@ func (conn *Conn) isConnectedOnAllWay() (status guard.ConnStatus) {
 	})
 }
 
+// enableWgWatcherIfNeeded starts a fresh watcher instance per connection attempt, so its
+// lifecycle stays bound to conn.mu and enable/disable can't race an old goroutine's shutdown.
+// Caller must hold conn.mu.
 func (conn *Conn) enableWgWatcherIfNeeded(enabledTime time.Time) {
-	if !conn.wgWatcher.PrepareInitialHandshake() {
+	if conn.wgWatcher != nil {
 		return
 	}
 
+	watcher := NewWGWatcher(conn.Log, conn.config.WgConfig.WgInterface, conn.config.Key, conn.dumpState)
+	watcher.PrepareInitialHandshake()
+
 	wgWatcherCtx, wgWatcherCancel := context.WithCancel(conn.ctx)
+	conn.wgWatcher = watcher
 	conn.wgWatcherCancel = wgWatcherCancel
+
 	conn.wgWatcherWg.Add(1)
 	go func() {
 		defer conn.wgWatcherWg.Done()
-		conn.wgWatcher.EnableWgWatcher(wgWatcherCtx, enabledTime, conn.onWGDisconnected, conn.onWGHandshakeSuccess, conn.onWGCheckSuccess)
+		onDisconnected := func() { conn.onWGDisconnected(wgWatcherCtx) }
+		watcher.EnableWgWatcher(wgWatcherCtx, enabledTime, onDisconnected, conn.onWGHandshakeSuccess, conn.onWGCheckSuccess)
 	}()
 }
 
+// disableWgWatcherIfNeeded cancels and drops the watcher once no transport is active. It never
+// waits for the goroutine: the timeout path reentrantly calls back here under conn.mu, so
+// blocking would deadlock. Caller must hold conn.mu.
 func (conn *Conn) disableWgWatcherIfNeeded() {
-	if conn.currentConnPriority == conntype.None && conn.wgWatcherCancel != nil {
-		conn.wgWatcherCancel()
-		conn.wgWatcherCancel = nil
+	if conn.currentConnPriority != conntype.None || conn.wgWatcher == nil {
+		return
 	}
+	conn.wgWatcherCancel()
+	conn.wgWatcher = nil
+	conn.wgWatcherCancel = nil
 }
 
 func (conn *Conn) newProxy(remoteConn net.Conn) (wgproxy.Proxy, error) {
@@ -874,7 +888,9 @@ func (conn *Conn) resetEndpoint() {
 		return
 	}
 	conn.Log.Infof("reset wg endpoint")
-	conn.wgWatcher.Reset()
+	if conn.wgWatcher != nil {
+		conn.wgWatcher.Reset()
+	}
 	if err := conn.endpointUpdater.RemoveEndpointAddress(); err != nil {
 		conn.Log.Warnf("failed to remove endpoint address before update: %v", err)
 	}

client/internal/peer/conn_test.go

@@ -339,20 +339,20 @@ func TestConn_onWGDisconnected_EscalatesToRosenpassReset(t *testing.T) {
 	conn := newWGTimeoutTestConn(true, &disconnected)
 
 	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
+		conn.onWGDisconnected(conn.ctx)
 	}
 	assert.Empty(t, disconnected, "escalation must not fire below the threshold")
 
-	conn.onWGDisconnected()
+	conn.onWGDisconnected(conn.ctx)
 	assert.Equal(t, []string{conn.config.WgConfig.RemoteKey}, disconnected,
 		"reaching the threshold must report the peer disconnected once")
 
 	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
+		conn.onWGDisconnected(conn.ctx)
 	}
 	assert.Len(t, disconnected, 1, "escalation must restart counting after firing")
 
-	conn.onWGDisconnected()
+	conn.onWGDisconnected(conn.ctx)
 	assert.Len(t, disconnected, 2, "continued timeouts must escalate again")
 }
 
@@ -364,12 +364,12 @@ func TestConn_onWGDisconnected_CheckSuccessResetsEscalation(t *testing.T) {
 	conn := newWGTimeoutTestConn(true, &disconnected)
 
 	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
+		conn.onWGDisconnected(conn.ctx)
 	}
 	conn.onWGCheckSuccess()
 
 	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
+		conn.onWGDisconnected(conn.ctx)
 	}
 	assert.Empty(t, disconnected, "handshake success must reset the timeout count")
 }
@@ -382,7 +382,7 @@ func TestConn_onWGDisconnected_NoEscalationWithoutRosenpass(t *testing.T) {
 	conn := newWGTimeoutTestConn(false, &disconnected)
 
 	for i := 0; i < wgTimeoutEscalationThreshold*3; i++ {
-		conn.onWGDisconnected()
+		conn.onWGDisconnected(conn.ctx)
 	}
 	assert.Empty(t, disconnected, "escalation must be limited to rosenpass connections")
 }

client/internal/peer/wg_watcher.go

@@ -3,7 +3,6 @@ package peer
 import (
 	"context"
 	"fmt"
-	"sync"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -24,14 +23,14 @@ type WGInterfaceStater interface {
 	GetStats() (map[string]configurer.WGStats, error)
 }
 
+// WGWatcher is single-shot: one instance per connection attempt, run once, then discarded.
+// Lifecycle is owned by Conn under conn.mu, so it keeps no "enabled" state to go stale.
 type WGWatcher struct {
 	log           *log.Entry
 	wgIfaceStater WGInterfaceStater
 	peerKey       string
 	stateDump     *stateDump
 
-	enabled   bool
-	muEnabled sync.Mutex
 	// initialHandshake is not thread-safe; never call PrepareInitialHandshake and EnableWgWatcher concurrently.
 	initialHandshake time.Time
 
@@ -48,25 +47,14 @@ func NewWGWatcher(log *log.Entry, wgIfaceStater WGInterfaceStater, peerKey strin
 	}
 }
 
-// PrepareInitialHandshake reserves the watcher and reads the peer's current WireGuard
-// handshake time. It must be called before the peer is (re)configured on the WireGuard
-// interface, so the captured baseline reflects the state prior to this connection attempt
-// instead of racing with that configuration. Returns ok=false if the watcher is already
-// running, in which case EnableWgWatcher must not be called.
-func (w *WGWatcher) PrepareInitialHandshake() (ok bool) {
-	w.muEnabled.Lock()
-	if w.enabled {
-		w.muEnabled.Unlock()
-		return false
-	}
-
+// PrepareInitialHandshake reads the peer's current WireGuard handshake time. It must be
+// called before the peer is (re)configured on the WireGuard interface, so the captured
+// baseline reflects the state prior to this connection attempt instead of racing with
+// that configuration.
+func (w *WGWatcher) PrepareInitialHandshake() {
 	w.log.Debugf("enable WireGuard watcher")
-	w.enabled = true
-	w.muEnabled.Unlock()
-
 	handshake, _ := w.wgState()
 	w.initialHandshake = handshake
-	return true
 }
 
 // EnableWgWatcher runs the WireGuard watcher loop using the handshake baseline captured by
@@ -76,10 +64,6 @@ func (w *WGWatcher) PrepareInitialHandshake() (ok bool) {
 // handshake, including the first.
 func (w *WGWatcher) EnableWgWatcher(ctx context.Context, enabledTime time.Time, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), onCheckSuccessFn func()) {
 	w.periodicHandshakeCheck(ctx, onDisconnectedFn, onHandshakeSuccessFn, onCheckSuccessFn, enabledTime, w.initialHandshake)
-
-	w.muEnabled.Lock()
-	w.enabled = false
-	w.muEnabled.Unlock()
 }
 
 // Reset signals the watcher that the WireGuard peer has been reset and a new
@@ -105,6 +89,7 @@ func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn
 		case <-timer.C:
 			handshake, ok := w.handshakeCheck(lastHandshake)
 			if !ok {
+				// early ctx cancel check return
 				if ctx.Err() != nil {
 					return
 				}
@@ -153,9 +138,9 @@ func (w *WGWatcher) handshakeCheck(lastHandshake time.Time) (*time.Time, bool) {
 
 	w.log.Tracef("previous handshake, handshake: %v, %v", lastHandshake, handshake)
 
-	// the current know handshake did not change
+	// the current known handshake did not change
 	if handshake.Equal(lastHandshake) {
-		w.log.Warnf("WireGuard handshake timed out: %v", handshake)
+		w.log.Warnf("WireGuard handshake not updated: %v", handshake)
 		return nil, false
 	}
 

client/internal/peer/wg_watcher_test.go

@@ -7,7 +7,6 @@ import (
 	"time"
 
 	log "github.com/sirupsen/logrus"
-	"github.com/stretchr/testify/require"
 
 	"github.com/netbirdio/netbird/client/iface/configurer"
 )
@@ -62,7 +61,7 @@ func TestWGWatcher_CheckSuccessCallback(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
-	require.True(t, watcher.PrepareInitialHandshake())
+	watcher.PrepareInitialHandshake()
 
 	firstHandshake := make(chan struct{}, 1)
 	checkSuccess := make(chan struct{}, 1)
@@ -101,8 +100,7 @@ func TestWGWatcher_EnableWgWatcher(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
-	ok := watcher.PrepareInitialHandshake()
-	require.True(t, ok, "watcher should not be enabled yet")
+	watcher.PrepareInitialHandshake()
 
 	onDisconnected := make(chan struct{}, 1)
 	go watcher.EnableWgWatcher(ctx, time.Now(), func() {
@@ -132,8 +130,7 @@ func TestWGWatcher_ReEnable(t *testing.T) {
 	watcher := NewWGWatcher(mlog, mocWgIface, "", newStateDump("peer", mlog, &Status{}))
 
 	ctx, cancel := context.WithCancel(context.Background())
-	ok := watcher.PrepareInitialHandshake()
-	require.True(t, ok, "watcher should not be enabled yet")
+	watcher.PrepareInitialHandshake()
 
 	wg := &sync.WaitGroup{}
 	wg.Add(1)
@@ -149,8 +146,7 @@ func TestWGWatcher_ReEnable(t *testing.T) {
 	ctx, cancel = context.WithCancel(context.Background())
 	defer cancel()
 
-	ok = watcher.PrepareInitialHandshake()
-	require.True(t, ok, "watcher should be re-enabled after the previous run stopped")
+	watcher.PrepareInitialHandshake()
 
 	onDisconnected := make(chan struct{}, 1)
 	go watcher.EnableWgWatcher(ctx, time.Now(), func() {

client/mdm/canonical_loaders.go

@@ -22,6 +22,7 @@ var allKeys = []string{
 	KeyDisableMetricsCollection,
 	KeyAllowServerSSH,
 	KeyDisableAutoConnect,
+	KeyDisableAutostart,
 	KeyPreSharedKey,
 	KeyRosenpassEnabled,
 	KeyRosenpassPermissive,

client/mdm/policy.go

@@ -20,10 +20,10 @@ import (
 // names (lowerCamelCase) so the daemon can map a Policy key directly to a
 // configuration field.
 const (
-	KeyManagementURL            = "managementURL"
-	KeyDisableUpdateSettings    = "disableUpdateSettings"
-	KeyDisableProfiles          = "disableProfiles"
-	KeyDisableNetworks          = "disableNetworks"
+	KeyManagementURL         = "managementURL"
+	KeyDisableUpdateSettings = "disableUpdateSettings"
+	KeyDisableProfiles       = "disableProfiles"
+	KeyDisableNetworks       = "disableNetworks"
 	// KeyDisableAdvancedView gates the advanced-view section in the
 	// upcoming UI revision. UI-only: NOT stored on Config, not
 	// applied by applyMDMPolicy, not rejectable via SetConfig. The
@@ -37,10 +37,16 @@ const (
 	KeyDisableMetricsCollection = "disableMetricsCollection"
 	KeyAllowServerSSH           = "allowServerSSH"
 	KeyDisableAutoConnect       = "disableAutoConnect"
-	KeyPreSharedKey             = "preSharedKey"
-	KeyRosenpassEnabled         = "rosenpassEnabled"
-	KeyRosenpassPermissive      = "rosenpassPermissive"
-	KeyWireguardPort            = "wireguardPort"
+	// KeyDisableAutostart suppresses the GUI's fresh-install
+	// launch-on-login default and marks the Settings toggle as
+	// MDM-managed. UI-only: NOT stored on Config and not applied by
+	// applyMDMPolicy; the GUI reads it directly and it appears in
+	// GetConfigResponse.mDMManagedFields when set.
+	KeyDisableAutostart    = "disableAutostart"
+	KeyPreSharedKey        = "preSharedKey"
+	KeyRosenpassEnabled    = "rosenpassEnabled"
+	KeyRosenpassPermissive = "rosenpassPermissive"
+	KeyWireguardPort       = "wireguardPort"
 
 	// Split tunnel is modeled as a single conceptual policy with two
 	// registry/plist values. KeySplitTunnelMode is the discriminator

client/server/server.go

@@ -181,7 +181,7 @@ func (s *Server) Start() error {
 		log.Warnf("failed to redirect stderr: %v", err)
 	}
 
-	if err := restoreResidualState(s.rootCtx, s.profileManager.GetStatePath()); err != nil {
+	if err := RestoreResidualState(s.rootCtx, s.profileManager.GetStatePath()); err != nil {
 		log.Warnf(errRestoreResidualState, err)
 	}
 
@@ -551,7 +551,7 @@ func (s *Server) Login(callerCtx context.Context, msg *proto.LoginRequest) (*pro
 	s.actCancel = cancel
 	s.mutex.Unlock()
 
-	if err := restoreResidualState(s.rootCtx, s.profileManager.GetStatePath()); err != nil {
+	if err := RestoreResidualState(s.rootCtx, s.profileManager.GetStatePath()); err != nil {
 		log.Warnf(errRestoreResidualState, err)
 	}
 
@@ -858,7 +858,7 @@ func (s *Server) Up(callerCtx context.Context, msg *proto.UpRequest) (*proto.UpR
 
 		return s.waitForUp(callerCtx)
 	}
-	if err := restoreResidualState(callerCtx, s.profileManager.GetStatePath()); err != nil {
+	if err := RestoreResidualState(callerCtx, s.profileManager.GetStatePath()); err != nil {
 		log.Warnf(errRestoreResidualState, err)
 	}
 

client/server/state.go

@@ -46,7 +46,7 @@ func (s *Server) CleanState(ctx context.Context, req *proto.CleanStateRequest) (
 
 	if req.All {
 		// Reuse existing cleanup logic for all states
-		if err := restoreResidualState(ctx, statePath); err != nil {
+		if err := RestoreResidualState(ctx, statePath); err != nil {
 			return nil, status.Errorf(codes.Internal, "failed to clean all states: %v", err)
 		}
 
@@ -113,9 +113,9 @@ func (s *Server) DeleteState(ctx context.Context, req *proto.DeleteStateRequest)
 	}, nil
 }
 
-// restoreResidualState checks if the client was not shut down in a clean way and restores residual if required.
+// RestoreResidualState checks if the client was not shut down in a clean way and restores residual if required.
 // Otherwise, we might not be able to connect to the management server to retrieve new config.
-func restoreResidualState(ctx context.Context, statePath string) error {
+func RestoreResidualState(ctx context.Context, statePath string) error {
 	if statePath == "" {
 		return nil
 	}

client/status/status.go

@@ -746,6 +746,8 @@ func ToProtoFullStatus(fullStatus peer.FullStatus) *proto.FullStatus {
 		pbFullStatus.DnsServers = append(pbFullStatus.DnsServers, pbDnsState)
 	}
 
+	pbFullStatus.Events = fullStatus.Events
+
 	return &pbFullStatus
 }
 

client/ui/autostart_default.go

@@ -0,0 +1,107 @@
+//go:build !android && !ios && !freebsd && !js
+
+package main
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
+	"github.com/netbirdio/netbird/client/ui/preferences"
+	"github.com/netbirdio/netbird/client/ui/services"
+)
+
+// autostartDefaultState carries the guard inputs of the one-time autostart
+// default decision so the decision itself stays a pure, testable function.
+type autostartDefaultState struct {
+	supported    bool
+	mdmDisabled  bool
+	priorInstall bool
+}
+
+// shouldEnableAutostartDefault applies the first-run guards in order and
+// returns whether autostart may be enabled, plus the reason when it may not.
+func shouldEnableAutostartDefault(s autostartDefaultState) (bool, string) {
+	switch {
+	case !s.supported:
+		return false, "autostart not supported on this platform"
+	case s.mdmDisabled:
+		return false, "autostart disabled by MDM policy"
+	case s.priorInstall:
+		return false, "existing NetBird installation"
+	}
+	return true, ""
+}
+
+// autostartDisabledByMDM reports whether the MDM policy manages the
+// disableAutostart key in a way that must suppress the default. An
+// unparseable managed value is treated as disabled to stay on the safe side.
+func autostartDisabledByMDM(policy *mdm.Policy) bool {
+	if !policy.HasKey(mdm.KeyDisableAutostart) {
+		return false
+	}
+	disabled, ok := policy.GetBool(mdm.KeyDisableAutostart)
+	return !ok || disabled
+}
+
+// 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.
+func netbirdFootprintExists() bool {
+	candidates := []string{
+		profilemanager.DefaultConfigPath,
+		filepath.Join(profilemanager.DefaultConfigPathDir, "config.json"),
+		filepath.Join(profilemanager.DefaultConfigPathDir, "state.json"),
+	}
+	for _, path := range candidates {
+		if path != "" && fileExists(path) {
+			return true
+		}
+	}
+	return false
+}
+
+// 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
+// Settings is never overridden: the marker guarantees at-most-once, ever.
+func applyAutostartDefault(ctx context.Context, autostart *services.Autostart, prefs *preferences.Store, prefsFileExisted bool) {
+	priorFootprint := netbirdFootprintExists() || prefsFileExisted
+
+	if prefs.Get().AutostartInitialized {
+		return
+	}
+	if err := prefs.SetAutostartInitialized(true); err != nil {
+		log.Warnf("persist autostart marker, skipping autostart default: %v", err)
+		return
+	}
+
+	state := autostartDefaultState{
+		supported:    autostart.Supported(ctx),
+		mdmDisabled:  autostartDisabledByMDM(mdm.LoadPolicy()),
+		priorInstall: priorFootprint,
+	}
+	enable, reason := shouldEnableAutostartDefault(state)
+	if !enable {
+		log.Debugf("skipping autostart default: %s", reason)
+		return
+	}
+
+	if err := autostart.SetEnabled(ctx, true); err != nil {
+		log.Warnf("enable autostart on fresh install: %v", err)
+		return
+	}
+	log.Info("autostart enabled by default on fresh install")
+}
+
+// fileExists reports whether path exists.
+func fileExists(path string) bool {
+	_, err := os.Stat(path)
+	return err == nil
+}

client/ui/autostart_default_test.go

@@ -0,0 +1,125 @@
+//go:build !android && !ios && !freebsd && !js
+
+package main
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+
+	"github.com/netbirdio/netbird/client/mdm"
+)
+
+func TestShouldEnableAutostartDefault(t *testing.T) {
+	allPass := autostartDefaultState{
+		supported:    true,
+		mdmDisabled:  false,
+		priorInstall: false,
+	}
+
+	tests := []struct {
+		name       string
+		mutate     func(*autostartDefaultState)
+		wantEnable bool
+		wantReason string
+	}{
+		{
+			name:       "fresh install with all guards passing enables",
+			mutate:     func(*autostartDefaultState) {},
+			wantEnable: true,
+		},
+		{
+			name:       "unsupported platform skips",
+			mutate:     func(s *autostartDefaultState) { s.supported = false },
+			wantReason: "autostart not supported on this platform",
+		},
+		{
+			name:       "MDM disable skips",
+			mutate:     func(s *autostartDefaultState) { s.mdmDisabled = true },
+			wantReason: "autostart disabled by MDM policy",
+		},
+		{
+			name:       "existing installation (upgrade) skips",
+			mutate:     func(s *autostartDefaultState) { s.priorInstall = true },
+			wantReason: "existing NetBird installation",
+		},
+		{
+			name: "unsupported wins over every other guard",
+			mutate: func(s *autostartDefaultState) {
+				s.supported = false
+				s.mdmDisabled = true
+				s.priorInstall = true
+			},
+			wantReason: "autostart not supported on this platform",
+		},
+		{
+			name: "MDM disable wins over prior install",
+			mutate: func(s *autostartDefaultState) {
+				s.mdmDisabled = true
+				s.priorInstall = true
+			},
+			wantReason: "autostart disabled by MDM policy",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			state := allPass
+			tc.mutate(&state)
+			enable, reason := shouldEnableAutostartDefault(state)
+			assert.Equal(t, tc.wantEnable, enable, "enable decision should match for state %+v", state)
+			assert.Equal(t, tc.wantReason, reason, "skip reason should identify the failing guard")
+		})
+	}
+}
+
+func TestAutostartDisabledByMDM(t *testing.T) {
+	tests := []struct {
+		name   string
+		values map[string]any
+		want   bool
+	}{
+		{
+			name:   "empty policy does not disable",
+			values: nil,
+			want:   false,
+		},
+		{
+			name:   "unrelated managed keys do not disable",
+			values: map[string]any{mdm.KeyDisableAutoConnect: true},
+			want:   false,
+		},
+		{
+			name:   "disableAutostart true disables",
+			values: map[string]any{mdm.KeyDisableAutostart: true},
+			want:   true,
+		},
+		{
+			name:   "disableAutostart registry DWORD 1 disables",
+			values: map[string]any{mdm.KeyDisableAutostart: int64(1)},
+			want:   true,
+		},
+		{
+			name:   "disableAutostart string true disables",
+			values: map[string]any{mdm.KeyDisableAutostart: "true"},
+			want:   true,
+		},
+		{
+			name:   "disableAutostart explicit false allows",
+			values: map[string]any{mdm.KeyDisableAutostart: false},
+			want:   false,
+		},
+		{
+			name:   "unparseable managed value is treated as disabled",
+			values: map[string]any{mdm.KeyDisableAutostart: "not-a-bool"},
+			want:   true,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := autostartDisabledByMDM(mdm.NewPolicy(tc.values))
+			assert.Equal(t, tc.want, got, "MDM disable decision should match for values %v", tc.values)
+		})
+	}
+}

client/ui/frontend/src/lib/connection.ts

@@ -51,7 +51,14 @@ async function runSsoLogin(
     if (uri) await openBrowserLoginUri(uri);
 
     const cancelPromise = buildSsoCancelPromise(state, signal);
-    const waitPromise = Connection.WaitSSOLogin({ userCode: result.userCode, hostname: "" });
+    // Combine wait + up in Go so the connection comes up the moment SSO
+    // completes. During SSO the tray window is hidden and the webview is
+    // suspended, so a frontend-driven Up (a promise continuation) would not
+    // fire until the user woke the window (e.g. hovering the tray icon).
+    const waitPromise = Connection.WaitSSOLoginAndUp(
+        { userCode: result.userCode, hostname: "" },
+        { profileName: "", username: "" },
+    );
 
     try {
         await Promise.race([waitPromise, cancelPromise]);
@@ -89,13 +96,13 @@ export async function startConnection(onSettled?: () => void, signal?: AbortSign
         if (signal?.aborted) state.cancelled = true;
 
         if (!state.cancelled && result.needsSsoLogin) {
+            // runSsoLogin brings the connection up in Go once SSO completes.
             await runSsoLogin(result, state, signal);
-        }
-
-        if (!state.cancelled && signal?.aborted) state.cancelled = true;
-
-        if (!state.cancelled) {
-            await Connection.Up({ profileName: "", username: "" });
+        } else {
+            if (!state.cancelled && signal?.aborted) state.cancelled = true;
+            if (!state.cancelled) {
+                await Connection.Up({ profileName: "", username: "" });
+            }
         }
     } catch (e) {
         WindowManager.CloseBrowserLogin().catch(console.error);

client/ui/main.go

@@ -197,6 +197,9 @@ func main() {
 		// daemon may keep the main window from showing, so the OS toast is the
 		// only reliable signal the user gets.
 		go notifyIfDaemonOutdated(compat, notifier, localizer)
+		// One-time launch-on-login default for fresh installs; gated by the
+		// NetBird footprint check, MDM policy, and the persisted marker.
+		go applyAutostartDefault(context.Background(), services.NewAutostart(app.Autostart), prefStore, prefStore.ExistedAtLoad())
 	})
 
 	if err := app.Run(); err != nil {

client/ui/preferences/store.go

@@ -54,6 +54,10 @@ type UIPreferences struct {
 	Language            i18n.LanguageCode `json:"language"`
 	ViewMode            ViewMode          `json:"viewMode"`
 	OnboardingCompleted bool              `json:"onboardingCompleted"`
+	// AutostartInitialized records that the one-time autostart default
+	// decision has run for this OS user. It only ever transitions to true
+	// and is never reset, so the default-on flow runs at most once, ever.
+	AutostartInitialized bool `json:"autostartInitialized"`
 }
 
 // LanguageValidator rejects SetLanguage inputs with no shipped bundle.
@@ -72,8 +76,9 @@ type Emitter interface {
 type Store struct {
 	path string
 
-	mu      sync.RWMutex
-	current UIPreferences
+	mu            sync.RWMutex
+	current       UIPreferences
+	existedAtLoad bool
 
 	subsMu sync.Mutex
 	subs   []chan UIPreferences
@@ -157,6 +162,27 @@ func (s *Store) SetOnboardingCompleted(done bool) error {
 	return nil
 }
 
+// SetAutostartInitialized persists the one-time autostart decision marker.
+// No-op if unchanged.
+func (s *Store) SetAutostartInitialized(done bool) error {
+	s.mu.Lock()
+	if s.current.AutostartInitialized == done {
+		s.mu.Unlock()
+		return nil
+	}
+	next := s.current
+	next.AutostartInitialized = done
+	if err := s.persistLocked(next); err != nil {
+		s.mu.Unlock()
+		return fmt.Errorf("persist preferences: %w", err)
+	}
+	s.current = next
+	s.mu.Unlock()
+
+	s.broadcast(next)
+	return nil
+}
+
 // SetLanguage validates, persists, and broadcasts. No-op if unchanged.
 func (s *Store) SetLanguage(lang i18n.LanguageCode) error {
 	if lang == "" {
@@ -206,13 +232,29 @@ func (s *Store) Subscribe() (<-chan UIPreferences, func()) {
 	return ch, unsubscribe
 }
 
+// ExistedAtLoad reports whether the backing preferences file was present on
+// disk when the store loaded. It distinguishes a user who ran a prior GUI
+// version from a brand-new OS user with no preferences yet.
+func (s *Store) ExistedAtLoad() bool {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	return s.existedAtLoad
+}
+
 // load reads the file into current. A missing file is not an error (the
 // in-memory default stands); malformed contents return an error.
 func (s *Store) load() error {
-	if _, err := os.Stat(s.path); errors.Is(err, os.ErrNotExist) {
-		return nil
+	if _, err := os.Stat(s.path); err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return nil
+		}
+		return fmt.Errorf("stat preferences: %w", err)
 	}
 
+	s.mu.Lock()
+	s.existedAtLoad = true
+	s.mu.Unlock()
+
 	var loaded UIPreferences
 	if _, err := util.ReadJson(s.path, &loaded); err != nil {
 		return err

client/ui/preferences/store_test.go

@@ -215,6 +215,46 @@ func TestStore_FileShapeIsJSON(t *testing.T) {
 	assert.Equal(t, i18n.LanguageCode("hu"), parsed.Language)
 }
 
+func TestStore_SetAutostartInitializedPersistsAcrossReload(t *testing.T) {
+	withTempConfigDir(t)
+	emitter := &recordingEmitter{}
+	s, err := NewStore(nil, emitter)
+	require.NoError(t, err)
+
+	assert.False(t, s.Get().AutostartInitialized, "marker must default to false when no file is on disk")
+
+	require.NoError(t, s.SetAutostartInitialized(true))
+	assert.True(t, s.Get().AutostartInitialized, "Get should reflect the persisted marker")
+	require.Len(t, emitter.calledWith(EventPreferencesChanged), 1, "first marker write should broadcast")
+
+	// Re-setting the same value must be a no-op: no disk write, no broadcast.
+	require.NoError(t, s.SetAutostartInitialized(true))
+	assert.Len(t, emitter.calledWith(EventPreferencesChanged), 1, "idempotent marker write should not broadcast again")
+
+	// A fresh Store (new GUI launch) must see the marker so the autostart
+	// default decision never runs twice.
+	reloaded, err := NewStore(nil, nil)
+	require.NoError(t, err)
+	assert.True(t, reloaded.Get().AutostartInitialized, "marker must survive a reload from disk")
+}
+
+func TestStore_ExistedAtLoad(t *testing.T) {
+	withTempConfigDir(t)
+
+	// Brand-new OS user: no preferences file on disk yet.
+	fresh, err := NewStore(nil, nil)
+	require.NoError(t, err)
+	assert.False(t, fresh.ExistedAtLoad(), "ExistedAtLoad must be false when no file is on disk")
+
+	// Persisting a value writes the file to disk.
+	require.NoError(t, fresh.SetLanguage("en"))
+
+	// A subsequent GUI launch reopens the now-present file.
+	reopened, err := NewStore(nil, nil)
+	require.NoError(t, err)
+	assert.True(t, reopened.ExistedAtLoad(), "ExistedAtLoad must be true after the store has persisted and is reopened")
+}
+
 func TestStore_ErrUnsupportedSentinel(t *testing.T) {
 	// Verifies callers can match on the sentinel error rather than parsing
 	// strings — protects against accidental %v -> %w changes that would

client/ui/services/connection.go

@@ -35,7 +35,7 @@ type LoginResult struct {
 	VerificationURIComplete string `json:"verificationUriComplete"`
 }
 
-// WaitSSOParams are the inputs to WaitSSOLogin.
+// WaitSSOParams are the inputs to waitSSOLogin.
 type WaitSSOParams struct {
 	UserCode string `json:"userCode"`
 	Hostname string `json:"hostname"`
@@ -125,23 +125,6 @@ func (s *Connection) Login(ctx context.Context, p LoginParams) (LoginResult, err
 	}, nil
 }
 
-func (s *Connection) WaitSSOLogin(ctx context.Context, p WaitSSOParams) (string, error) {
-	cli, err := s.conn.Client()
-	if err != nil {
-		return "", err
-	}
-	log.Infof("waiting for SSO login to complete")
-	resp, err := cli.WaitSSOLogin(ctx, &proto.WaitSSOLoginRequest{
-		UserCode: p.UserCode,
-		Hostname: p.Hostname,
-	})
-	if err != nil {
-		return "", s.classifyDaemonError(err)
-	}
-	log.Infof("SSO login completed, daemon reported success")
-	return resp.GetEmail(), nil
-}
-
 func (s *Connection) Up(ctx context.Context, p UpParams) error {
 	cli, err := s.conn.Client()
 	if err != nil {
@@ -162,6 +145,27 @@ func (s *Connection) Up(ctx context.Context, p UpParams) error {
 	return nil
 }
 
+// WaitSSOLoginAndUp blocks until the SSO login completes and then brings the
+// connection up, both from the Go side. Keeping the post-login Up here rather
+// than as a frontend continuation is deliberate: during SSO the tray window is
+// hidden and the webview is suspended (macOS App Nap / hidden-window timer
+// throttling), so a frontend-driven Up would not run until the user woke the
+// window (e.g. by hovering the tray icon). Doing it in Go connects the moment
+// the daemon reports SSO success. Returns the authenticated user's email.
+func (s *Connection) WaitSSOLoginAndUp(ctx context.Context, wait WaitSSOParams, up UpParams) (string, error) {
+	email, err := s.waitSSOLogin(ctx, wait)
+	if err != nil {
+		return "", err
+	}
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
+	if err := s.Up(ctx, up); err != nil {
+		return "", err
+	}
+	return email, nil
+}
+
 func (s *Connection) Down(ctx context.Context) error {
 	cli, err := s.conn.Client()
 	if err != nil {
@@ -221,6 +225,26 @@ func (s *Connection) Logout(ctx context.Context, p LogoutParams) error {
 	return nil
 }
 
+// waitSSOLogin blocks until the daemon reports the SSO login result and returns
+// the authenticated user's email. It is unexported because the frontend drives
+// SSO through the exported WaitSSOLoginAndUp.
+func (s *Connection) waitSSOLogin(ctx context.Context, p WaitSSOParams) (string, error) {
+	cli, err := s.conn.Client()
+	if err != nil {
+		return "", err
+	}
+	log.Infof("waiting for SSO login to complete")
+	resp, err := cli.WaitSSOLogin(ctx, &proto.WaitSSOLoginRequest{
+		UserCode: p.UserCode,
+		Hostname: p.Hostname,
+	})
+	if err != nil {
+		return "", s.classifyDaemonError(err)
+	}
+	log.Infof("SSO login completed, daemon reported success")
+	return resp.GetEmail(), nil
+}
+
 // classifyDaemonError maps a gRPC error to a localised ClientError.
 func (s *Connection) classifyDaemonError(err error) *ClientError {
 	return s.classifier.classify(err)

client/ui/services/settings.go

@@ -20,11 +20,12 @@ type MDMFields struct {
 	DisableServerRoutes      bool   `json:"disableServerRoutes"`
 	AllowServerSSH           *bool  `json:"allowServerSSH"`
 	DisableAutoConnect       bool   `json:"disableAutoConnect"`
+	DisableAutostart         bool   `json:"disableAutostart"`
 	BlockInbound             bool   `json:"blockInbound"`
 	DisableMetricsCollection bool   `json:"disableMetricsCollection"`
 	SplitTunnelMode          bool   `json:"splitTunnelMode"`
 	SplitTunnelApps          bool   `json:"splitTunnelApps"`
-	DisableAdvancedView bool `json:"disableAdvancedView"`
+	DisableAdvancedView      bool   `json:"disableAdvancedView"`
 }
 
 type Features struct {

e2e/agentnetwork/chat_test.go

@@ -91,7 +91,7 @@ func availableProviders() []providerCase {
 		if region == "" {
 			region = "us-east-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.WireMessages})
+		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})
 	}
 	return ps
 }
@@ -224,9 +224,12 @@ func TestProvidersMatrix(t *testing.T) {
 				var c int
 				var b string
 				var cerr error
-				if pc.kind == harness.WireVertex {
+				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)
-				} else {
+				case harness.WireBedrock:
+					c, b, cerr = cl.Bedrock(ctx, settings.Endpoint, proxyIP, pc.model, "Reply with exactly: pong", sessionID)
+				default:
 					c, b, cerr = cl.Chat(ctx, settings.Endpoint, proxyIP, pc.kind, pc.model, "Reply with exactly: pong", sessionID)
 				}
 				if cerr == nil {

e2e/agentnetwork/guardrail_test.go

@@ -0,0 +1,168 @@
+//go:build e2e
+
+package agentnetwork
+
+import (
+	"context"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/e2e/harness"
+	"github.com/netbirdio/netbird/shared/management/http/api"
+)
+
+// 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).
+func catalogModel(pc providerCase) string {
+	switch pc.kind {
+	case harness.WireBedrock:
+		return strings.TrimPrefix(pc.model, "us.")
+	case harness.WireVertex:
+		return strings.SplitN(pc.model, "@", 2)[0]
+	default:
+		return pc.model
+	}
+}
+
+// disallowedModel returns a valid-shaped model id for the provider that is NOT
+// the configured/allowed one, so the guardrail must reject it before the
+// request ever reaches the upstream.
+func disallowedModel(pc providerCase) string {
+	switch pc.kind {
+	case harness.WireBedrock:
+		return "us.anthropic.claude-opus-4-8"
+	case harness.WireVertex:
+		return "claude-opus-4-8@20250101"
+	default:
+		return "unlisted-model"
+	}
+}
+
+// sendModel drives one request for the given model through the provider's native
+// wire shape and returns the HTTP status.
+func sendModel(ctx context.Context, t *testing.T, cl *harness.Client, endpoint, proxyIP string, pc providerCase, model string) int {
+	t.Helper()
+	var code int
+	var err error
+	switch pc.kind {
+	case harness.WireBedrock:
+		code, _, err = cl.Bedrock(ctx, endpoint, proxyIP, model, "Reply with exactly: pong", "")
+	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", "")
+	}
+	require.NoError(t, err, "request must reach the proxy for %s", pc.name)
+	return code
+}
+
+// TestModelAllowlistEnforced provisions a Model Allowlist guardrail limiting each
+// path-routed provider (Bedrock, Vertex) to its configured model, then drives
+// requests over the tunnel: the allowed model returns 200 while a model outside
+// the allowlist is denied 403 by the guardrail before it reaches the upstream.
+// This is the coverage missing for #6751 — the model for these providers travels
+// in the URL path, and the allowlist must be enforced there.
+func TestModelAllowlistEnforced(t *testing.T) {
+	var providers []providerCase
+	for _, pc := range availableProviders() {
+		if pc.kind == harness.WireBedrock || pc.kind == harness.WireVertex {
+			providers = append(providers, pc)
+		}
+	}
+	if len(providers) == 0 {
+		t.Skip("no path-routed provider keys set (AWS_BEARER_TOKEN_BEDROCK / GOOGLE_VERTEX_*); source ~/.llm-keys")
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Minute)
+	defer cancel()
+
+	grp, err := srv.API().Groups.Create(ctx, api.PostApiGroupsJSONRequestBody{Name: "e2e-allowlist"})
+	require.NoError(t, err, "create group")
+	t.Cleanup(func() { _ = srv.API().Groups.Delete(context.Background(), grp.Id) })
+
+	ephemeral := false
+	sk, err := srv.API().SetupKeys.Create(ctx, api.PostApiSetupKeysJSONRequestBody{
+		Name:       "e2e-allowlist-client",
+		Type:       "reusable",
+		ExpiresIn:  86400,
+		UsageLimit: 0,
+		AutoGroups: []string{grp.Id},
+		Ephemeral:  &ephemeral,
+	})
+	require.NoError(t, err, "mint setup key")
+
+	// Providers with their configured (allowed) models; the first bootstraps the cluster.
+	ids := make([]string, 0, len(providers))
+	allowed := make([]string, 0, len(providers))
+	for i, pc := range providers {
+		req := providerRequest(pc)
+		if i == 0 {
+			req.BootstrapCluster = ptr(harness.AgentNetworkCluster)
+		}
+		prov, perr := srv.CreateProvider(ctx, req)
+		require.NoError(t, perr, "create provider %s", pc.name)
+		id := prov.Id
+		ids = append(ids, id)
+		allowed = append(allowed, catalogModel(pc))
+		t.Cleanup(func() { _ = srv.DeleteProvider(context.Background(), id) })
+	}
+
+	// Guardrail allowlisting exactly the configured models.
+	var gr api.AgentNetworkGuardrailRequest
+	gr.Name = "e2e-allowlist"
+	gr.Checks.ModelAllowlist.Enabled = true
+	gr.Checks.ModelAllowlist.Models = allowed
+	guard, err := srv.CreateGuardrail(ctx, gr)
+	require.NoError(t, err, "create guardrail")
+	t.Cleanup(func() { _ = srv.DeleteGuardrail(context.Background(), guard.Id) })
+
+	enabled := true
+	pol, err := srv.CreatePolicy(ctx, api.AgentNetworkPolicyRequest{
+		Name:                   "e2e-allowlist",
+		Enabled:                &enabled,
+		SourceGroups:           []string{grp.Id},
+		DestinationProviderIds: ids,
+		GuardrailIds:           &[]string{guard.Id},
+	})
+	require.NoError(t, err, "create policy")
+	t.Cleanup(func() { _ = srv.DeletePolicy(context.Background(), pol.Id) })
+
+	settings, err := srv.GetSettings(ctx)
+	require.NoError(t, err, "read settings for endpoint")
+	require.NotEmpty(t, settings.Endpoint, "agent-network endpoint must be assigned")
+
+	proxyToken, err := srv.CreateProxyTokenCLI(ctx, "e2e-proxy-allowlist")
+	require.NoError(t, err, "mint proxy token via CLI")
+	px, err := harness.StartProxy(ctx, srv, proxyToken)
+	require.NoError(t, err, "start proxy")
+	t.Cleanup(func() { _ = px.Terminate(context.Background()) })
+
+	cl, err := harness.StartClient(ctx, srv, sk.Key)
+	require.NoError(t, err, "start client")
+	t.Cleanup(func() { _ = cl.Terminate(context.Background()) })
+
+	require.NoError(t, cl.WaitConnected(ctx, 90*time.Second), "client must connect to management")
+	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
+		t.Run(pc.name, func(t *testing.T) {
+			// The admin's allowlisted model is served end to end.
+			assert.Equal(t, 200, sendModel(ctx, t, cl, settings.Endpoint, proxyIP, pc, pc.model),
+				"allowlisted model must be permitted for %s", pc.name)
+			// A model outside the allowlist is rejected by the guardrail (before
+			// the upstream), regardless of whether it is a real catalog model.
+			assert.Equal(t, 403, sendModel(ctx, t, cl, settings.Endpoint, proxyIP, pc, disallowedModel(pc)),
+				"model outside the allowlist must be denied for %s", pc.name)
+		})
+	}
+}

e2e/harness/agentnetwork.go

@@ -107,6 +107,17 @@ func (c *Combined) DeletePolicy(ctx context.Context, id string) error {
 	return anDelete(ctx, c, "/api/agent-network/policies/"+id)
 }
 
+// CreateGuardrail creates an agent-network guardrail (e.g. a model allowlist)
+// that can then be attached to a policy via its GuardrailIds.
+func (c *Combined) CreateGuardrail(ctx context.Context, req api.AgentNetworkGuardrailRequest) (api.AgentNetworkGuardrail, error) {
+	return anRequest[api.AgentNetworkGuardrail](ctx, c, http.MethodPost, "/api/agent-network/guardrails", req)
+}
+
+// DeleteGuardrail removes a guardrail by id.
+func (c *Combined) DeleteGuardrail(ctx context.Context, id string) error {
+	return anDelete(ctx, c, "/api/agent-network/guardrails/"+id)
+}
+
 // GetSettings returns the account's agent-network settings row. It exists only
 // after the first provider create bootstraps it.
 func (c *Combined) GetSettings(ctx context.Context) (api.AgentNetworkSettings, error) {

e2e/harness/client.go

@@ -194,6 +194,11 @@ const (
 	// WireVertex is the Anthropic-on-Vertex rawPredict shape: the client posts
 	// the full Vertex model path and the proxy mints the SA OAuth token.
 	WireVertex = "vertex"
+	// WireBedrock is the native AWS Bedrock InvokeModel shape: the model id
+	// travels in the URL path (/model/{id}/invoke), not the body, so the proxy
+	// routes by path. This is what a Bedrock SDK client sends and the shape the
+	// model-allowlist guardrail must enforce.
+	WireBedrock = "bedrock"
 )
 
 // Chat issues a chat-completion POST to the agent-network endpoint over the
@@ -226,6 +231,17 @@ func (cl *Client) Vertex(ctx context.Context, endpoint, proxyIP, project, region
 	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(nil, sessionID))
 }
 
+// Bedrock issues a native AWS Bedrock InvokeModel POST over the tunnel. The
+// model id is carried in the request path (/model/{id}/invoke), so the proxy
+// routes by path; the body uses the bedrock anthropic_version rather than a
+// model field. A non-empty sessionID is sent as the universal x-session-id
+// 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)
+	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(nil, sessionID))
+}
+
 // withSessionID appends the x-session-id header when sessionID is non-empty.
 func withSessionID(headers []string, sessionID string) []string {
 	if sessionID == "" {

management/internals/modules/peers/ephemeral/manager/ephemeral.go

@@ -215,7 +215,7 @@ func (e *EphemeralManager) cleanup(ctx context.Context) {
 	}
 
 	for accountID, peerIDs := range peerIDsPerAccount {
-		log.WithContext(ctx).Tracef("cleanup: deleting %d ephemeral peers for account %s", len(peerIDs), accountID)
+		log.WithContext(ctx).Debugf("cleanup: deleting %d ephemeral peers for account %s: %s", len(peerIDs), accountID, peerIDs)
 		err := e.peersManager.DeletePeers(ctx, accountID, peerIDs, activity.SystemInitiator, true)
 		if err != nil {
 			log.WithContext(ctx).Errorf("failed to delete ephemeral peers: %s", err)

management/internals/modules/peers/manager.go

@@ -184,6 +184,8 @@ func (m *managerImpl) DeletePeers(ctx context.Context, accountID string, peerIDs
 				return err
 			}
 
+			log.WithContext(ctx).Debugf("DeletePeers: deleted peer %s", peerID)
+
 			if !(peer.ProxyMeta.Embedded || peer.Meta.KernelVersion == "wasm") {
 				eventsToStore = append(eventsToStore, func() {
 					m.accountManager.StoreEvent(ctx, userID, peer.ID, accountID, activity.PeerRemovedByUser, peer.EventMeta(dnsDomain))
@@ -224,30 +226,6 @@ func (m *managerImpl) CreateProxyPeer(ctx context.Context, accountID string, pee
 		return nil
 	}
 
-	// Dedupe stale embedded peer records for the same (account, cluster).
-	// The proxy generates a fresh WireGuard keypair on every startup
-	// (proxy/internal/roundtrip/netbird.go), so without this sweep the
-	// prior embedded peer would linger forever — holding its CGNAT IP
-	// allocation, polluting other peers' rosters, and (most visibly)
-	// leaving the synth DNS pointing at the dead address. The
-	// (account, cluster) tuple identifies "the embedded peer for this
-	// proxy instance at this cluster"; any record matching that tuple
-	// with a different pubkey is by definition stale and must go.
-	staleIDs, err := m.findStaleEmbeddedProxyPeers(ctx, accountID, cluster, peerKey)
-	if err != nil {
-		return fmt.Errorf("scan for stale embedded proxy peers: %w", err)
-	}
-	if len(staleIDs) > 0 {
-		// userID="" + checkConnected=false: the deletion is initiated
-		// by management itself on behalf of the freshly-registering
-		// proxy, not by an end user; the stale peer may still be
-		// marked Connected from its prior session, but its session is
-		// dead by definition (its key no longer exists).
-		if err := m.DeletePeers(ctx, accountID, staleIDs, "", false); err != nil {
-			return fmt.Errorf("delete stale embedded proxy peers %v: %w", staleIDs, err)
-		}
-	}
-
 	name := fmt.Sprintf("proxy-%s", xid.New().String())
 	newPeer := &peer.Peer{
 		Ephemeral: true,
@@ -273,29 +251,3 @@ func (m *managerImpl) CreateProxyPeer(ctx context.Context, accountID string, pee
 
 	return nil
 }
-
-// findStaleEmbeddedProxyPeers returns the peer IDs of embedded proxy peer
-// records in accountID that target the same cluster but carry a different
-// WireGuard pubkey than the freshly-registering one. Used by CreateProxyPeer
-// to garbage-collect stale records left behind when the proxy restarts with a
-// regenerated keypair.
-func (m *managerImpl) findStaleEmbeddedProxyPeers(ctx context.Context, accountID, cluster, newKey string) ([]string, error) {
-	account, err := m.store.GetAccount(ctx, accountID)
-	if err != nil {
-		return nil, err
-	}
-	var stale []string
-	for _, p := range account.Peers {
-		if p == nil || !p.ProxyMeta.Embedded {
-			continue
-		}
-		if p.ProxyMeta.Cluster != cluster {
-			continue
-		}
-		if p.Key == newKey {
-			continue
-		}
-		stale = append(stale, p.ID)
-	}
-	return stale, nil
-}

management/internals/shared/grpc/token_mgr.go

@@ -161,6 +161,8 @@ func (m *TimeBasedAuthSecretsManager) SetupRefresh(ctx context.Context, accountI
 		m.turnCancelMap[peerID] = turnCancel
 		go m.refreshTURNTokens(ctx, accountID, peerID, turnCancel)
 		log.WithContext(ctx).Debugf("starting TURN refresh for %s", peerID)
+	} else {
+		log.WithContext(ctx).Debugf("no TURN configuration, skipping TURN refresh for %s", peerID)
 	}
 
 	if m.relayCfg != nil {
@@ -168,6 +170,8 @@ func (m *TimeBasedAuthSecretsManager) SetupRefresh(ctx context.Context, accountI
 		m.relayCancelMap[peerID] = relayCancel
 		go m.refreshRelayTokens(ctx, accountID, peerID, relayCancel)
 		log.WithContext(ctx).Tracef("starting relay refresh for %s", peerID)
+	} else {
+		log.WithContext(ctx).Tracef("no relay configuration, skipping relay refresh for %s", peerID)
 	}
 }
 

management/server/agentnetwork_proxypeer_restart_test.go

@@ -1,199 +0,0 @@
-package server
-
-import (
-	"context"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-
-	"github.com/netbirdio/netbird/management/internals/modules/peers"
-	"github.com/netbirdio/netbird/management/internals/modules/agentnetwork"
-	agenttypes "github.com/netbirdio/netbird/management/internals/modules/agentnetwork/types"
-	nbpeer "github.com/netbirdio/netbird/management/server/peer"
-	"github.com/netbirdio/netbird/management/server/permissions"
-	"github.com/netbirdio/netbird/management/server/store"
-	"github.com/netbirdio/netbird/management/server/types"
-)
-
-// TestAgentNetwork_ProxyRestart_PropagatesNewPeerAndDropsStale is the no-mock
-// regression guard for the bug the user reported: restarting the proxy creates
-// a fresh embedded peer with a NEW WireGuard public key (the proxy generates
-// the keypair on every startup at proxy/internal/roundtrip/netbird.go:312).
-// The PRIOR embedded peer record is never deleted on management, so the
-// account accumulates a stale peer holding a stale CGNAT IP. Other peers
-// in the account either keep routing to the dead IP, or — if synth DNS
-// picks the wrong record — never see the new IP at all.
-//
-// What this test exercises (no mocks):
-//   - real SQLite test store
-//   - real DefaultAccountManager, network-map controller, peer-update channels
-//   - real peers.Manager.CreateProxyPeer path (the very method the proxy
-//     invokes over gRPC on every startup)
-//   - real agentnetwork.Manager + synth chain so the client receives a
-//     concrete DNS record that must point at the LATEST proxy peer.
-//
-// Pre-fix expected behavior (red): two embedded peers exist after the
-// "restart"; the synth DNS record points at the stale one; the client
-// receives an update reflecting the new peer but the old one lingers.
-// Post-fix expected behavior (green): exactly one embedded peer exists
-// after restart (with the new key) AND the client's network map carries
-// the synth DNS pointing at that new peer's CGNAT IP.
-func TestAgentNetwork_ProxyRestart_PropagatesNewPeerAndDropsStale(t *testing.T) {
-	am, updateManager, err := createManager(t)
-	require.NoError(t, err, "createManager must succeed")
-	ctx := context.Background()
-
-	const (
-		accountID   = "an-restart-acct"
-		adminUserID = "an-restart-admin"
-		groupAID    = "an-restart-grp-A"
-		clusterAddr = "eu.proxy.netbird.io"
-		clientKey   = "BhRPtynAAYRDy08+q4HTMsos8fs4plTP4NOSh7C1ry8="
-		// Two different proxy pubkeys — the "before" and "after" of a
-		// proxy-process restart with fresh-keypair generation.
-		proxyKey1 = "Aaaaa1aaaaYRDy08+q4HTMsos8fs4plTP4NOSh7C1ry8="
-		proxyKey2 = "Bbbbb2bbbbYRDy08+q4HTMsos8fs4plTP4NOSh7C1ry8="
-	)
-
-	// --- Account scaffold ---
-	account := newAccountWithId(ctx, accountID, adminUserID, "an-restart.test", "", "", false)
-	require.NoError(t, am.Store.SaveAccount(ctx, account))
-
-	clientPeer := &nbpeer.Peer{
-		Key:      clientKey,
-		Name:     "an-restart-client",
-		DNSLabel: "an-restart-client",
-		Meta:     nbpeer.PeerSystemMeta{Hostname: "an-restart-client", GoOS: "linux", WtVersion: "development"},
-	}
-	addedClient, _, _, _, err := am.AddPeer(ctx, "", "", adminUserID, clientPeer, false)
-	require.NoError(t, err, "AddPeer for client must succeed")
-	require.NoError(t, am.MarkPeerConnected(ctx, clientKey, accountID, time.Now().UnixNano(), &types.NetworkMap{}),
-		"MarkPeerConnected for the client peer must succeed (affected-peer fan-out skips disconnected peers)")
-
-	// Place the client in group A so the synth policy reaches it.
-	account, err = am.Store.GetAccount(ctx, accountID)
-	require.NoError(t, err)
-	account.Groups[groupAID] = &types.Group{ID: groupAID, Name: "groupA", Peers: []string{addedClient.ID}}
-	require.NoError(t, am.Store.SaveAccount(ctx, account), "SaveAccount must persist group A")
-
-	// --- Real peers + agent-network managers ---
-	permMgr := permissions.NewManager(am.Store)
-	peersMgr := peers.NewManager(am.Store, permMgr)
-	peersMgr.SetAccountManager(am)
-	peersMgr.SetNetworkMapController(am.networkMapController)
-	agentMgr := agentnetwork.NewManager(am.Store, permMgr, am, nil)
-
-	// Subscribe BEFORE any state-mutating call so we don't lose the update
-	// that contains the synth DNS record.
-	clientCh := updateManager.CreateChannel(ctx, addedClient.ID)
-	t.Cleanup(func() { updateManager.CloseChannel(ctx, addedClient.ID) })
-	drain(clientCh)
-
-	// --- First proxy startup: register peer key K1, then mark it
-	// connected. In production the proxy follows CreateProxyPeer with the
-	// regular sync stream which lands on MarkPeerConnected; the synth DNS
-	// path filters out peers that aren't Connected (types/account.go:323),
-	// so without this step no DNS record would be emitted.
-	require.NoError(t, peersMgr.CreateProxyPeer(ctx, accountID, proxyKey1, clusterAddr),
-		"first CreateProxyPeer (proxy startup) must succeed")
-
-	peer1ID, err := am.Store.GetPeerIDByKey(ctx, store.LockingStrengthNone, proxyKey1)
-	require.NoError(t, err, "proxy peer for K1 must be persisted after CreateProxyPeer")
-	require.NotEmpty(t, peer1ID)
-
-	require.NoError(t, am.MarkPeerConnected(ctx, proxyKey1, accountID, time.Now().UnixNano(), &types.NetworkMap{}),
-		"MarkPeerConnected for K1 must succeed")
-
-	account, err = am.Store.GetAccount(ctx, accountID)
-	require.NoError(t, err)
-	proxyIP1 := account.Peers[peer1ID].IP.String()
-	require.NotEmpty(t, proxyIP1, "K1 must have an assigned overlay IP")
-
-	// --- Provider + policy. CreateProvider / CreatePolicy trigger the
-	// agentnetwork reconcile which runs UpdateAccountPeers; the resulting
-	// NetworkMap delivered to the client carries the synth DNS record
-	// pointing at K1's IP. ---
-	provider, err := agentMgr.CreateProvider(ctx, adminUserID, &agenttypes.Provider{
-		AccountID:   accountID,
-		ProviderID:  "openai_api",
-		Name:        "openai-test",
-		UpstreamURL: "https://api.openai.com",
-		APIKey:      "sk-test-key",
-		Enabled:     true,
-		Models:      []agenttypes.ProviderModel{{ID: "gpt-5.4"}},
-	}, clusterAddr)
-	require.NoError(t, err, "CreateProvider must succeed")
-
-	_, err = agentMgr.CreatePolicy(ctx, adminUserID, &agenttypes.Policy{
-		AccountID:              accountID,
-		Name:                   "p1",
-		Enabled:                true,
-		SourceGroups:           []string{groupAID},
-		DestinationProviderIDs: []string{provider.ID},
-	})
-	require.NoError(t, err, "CreatePolicy must succeed")
-
-	settings, err := am.Store.GetAgentNetworkSettings(ctx, store.LockingStrengthNone, accountID)
-	require.NoError(t, err)
-	fqdn := settings.Endpoint()
-
-	rdata1 := awaitZoneRData(clientCh, clusterAddr, fqdn, true)
-	require.Equal(t, proxyIP1, rdata1,
-		"client must receive a synth DNS record pointing at K1's overlay IP after the synth path runs")
-	drain(clientCh)
-
-	// --- Proxy restart: NEW keypair K2, same account, same cluster ---
-	require.NoError(t, peersMgr.CreateProxyPeer(ctx, accountID, proxyKey2, clusterAddr),
-		"second CreateProxyPeer (proxy restart with fresh keypair) must succeed")
-
-	peer2ID, err := am.Store.GetPeerIDByKey(ctx, store.LockingStrengthNone, proxyKey2)
-	require.NoError(t, err, "proxy peer for K2 must be persisted after restart")
-	require.NotEmpty(t, peer2ID)
-
-	require.NoError(t, am.MarkPeerConnected(ctx, proxyKey2, accountID, time.Now().UnixNano(), &types.NetworkMap{}),
-		"MarkPeerConnected for K2 must succeed")
-
-	// In production the agent's sync stream pulls a fresh NetworkMap as
-	// part of its normal reconcile cadence; in this isolated test
-	// MarkPeerConnected's affected-peer fan-out can race the channel-side
-	// buffer in a way that swallows the synth-DNS-bearing update before
-	// our await reads it. Trigger an explicit account-wide fan-out so the
-	// assertion below tests what production actually delivers, not the
-	// in-test buffer race.
-	am.UpdateAccountPeers(ctx, accountID, types.UpdateReason{Resource: types.UpdateResourcePeer, Operation: types.UpdateOperationUpdate})
-
-	account, err = am.Store.GetAccount(ctx, accountID)
-	require.NoError(t, err)
-	proxyIP2 := account.Peers[peer2ID].IP.String()
-	require.NotEmpty(t, proxyIP2, "K2 must have an assigned overlay IP")
-	require.NotEqual(t, proxyIP1, proxyIP2, "K2 must get a different overlay IP than K1 (sanity)")
-
-	// CRITICAL ASSERTION 1: K1 must no longer be in the store. The SqlStore
-	// returns ("", nil) for a missing key rather than NotFound, so assert
-	// on the returned ID being empty.
-	staleID, err := am.Store.GetPeerIDByKey(ctx, store.LockingStrengthNone, proxyKey1)
-	require.NoError(t, err, "GetPeerIDByKey for a missing peer must not error")
-	assert.Empty(t, staleID,
-		"stale embedded proxy peer K1 must be removed when a new embedded peer registers for the same (account, cluster); pre-fix this assertion fails because management never cleans up the prior peer record")
-
-	// CRITICAL ASSERTION 2: exactly one embedded proxy peer remains, and it
-	// is K2.
-	account, err = am.Store.GetAccount(ctx, accountID)
-	require.NoError(t, err)
-	embeddedKeys := []string{}
-	for _, p := range account.Peers {
-		if p.ProxyMeta.Embedded {
-			embeddedKeys = append(embeddedKeys, p.Key)
-		}
-	}
-	assert.Equal(t, []string{proxyKey2}, embeddedKeys,
-		"after a proxy restart exactly one embedded proxy peer should remain — the one with the new key K2")
-
-	// CRITICAL ASSERTION 3: the synth DNS record the client receives now
-	// points at K2's IP, not K1's.
-	rdata2 := awaitZoneRData(clientCh, clusterAddr, fqdn, true)
-	assert.Equal(t, proxyIP2, rdata2,
-		"after proxy restart, the client's synth DNS record must point at the NEW embedded peer's IP, not the stale K1 IP")
-}

management/server/http/handlers/accounts/accounts_handler.go

@@ -295,6 +295,13 @@ func (h *handler) updateAccountRequestSettings(req api.PutApiAccountsAccountIdJS
 		}
 	}
 
+	if returnSettings.AgentNetworkOnly &&
+		(returnSettings.DashboardFeatures == nil ||
+			returnSettings.DashboardFeatures.AgentNetwork == nil ||
+			!*returnSettings.DashboardFeatures.AgentNetwork) {
+		return nil, status.Errorf(status.InvalidArgument, "agent network only mode requires dashboard_features.agent_network to be enabled")
+	}
+
 	return returnSettings, nil
 }
 

management/server/http/handlers/accounts/accounts_handler_test.go

@@ -288,7 +288,7 @@ func TestAccounts_AccountsHandler(t *testing.T) {
 			expectedBody:   true,
 			requestType:    http.MethodPut,
 			requestPath:    "/api/accounts/" + accountID,
-			requestBody:    bytes.NewBufferString("{\"settings\": {\"peer_login_expiration\": 15552000,\"peer_login_expiration_enabled\": true,\"agent_network_only\": true},\"onboarding\": {\"onboarding_flow_pending\": true,\"signup_form_pending\": true}}"),
+			requestBody:    bytes.NewBufferString("{\"settings\": {\"peer_login_expiration\": 15552000,\"peer_login_expiration_enabled\": true,\"agent_network_only\": true,\"dashboard_features\": {\"agent_network\": true}},\"onboarding\": {\"onboarding_flow_pending\": true,\"signup_form_pending\": true}}"),
 			expectedStatus: http.StatusOK,
 			expectedSettings: api.AccountSettings{
 				PeerLoginExpiration:             15552000,
@@ -305,13 +305,25 @@ func TestAccounts_AccountsHandler(t *testing.T) {
 				AutoUpdateVersion:               sr(""),
 				MetricsPushEnabled:              br(false),
 				AgentNetworkOnly:                br(true),
-				EmbeddedIdpEnabled:              br(false),
-				LocalAuthDisabled:               br(false),
-				LocalMfaEnabled:                 br(false),
+				DashboardFeatures: &api.AccountDashboardFeatures{
+					AgentNetwork: br(true),
+				},
+				EmbeddedIdpEnabled: br(false),
+				LocalAuthDisabled:  br(false),
+				LocalMfaEnabled:    br(false),
 			},
 			expectedArray: false,
 			expectedID:    accountID,
 		},
+		{
+			name:           "PutAccount fails enabling agent_network_only without dashboard_features",
+			expectedBody:   true,
+			requestType:    http.MethodPut,
+			requestPath:    "/api/accounts/" + accountID,
+			requestBody:    bytes.NewBufferString("{\"settings\": {\"peer_login_expiration\": 15552000,\"peer_login_expiration_enabled\": true,\"agent_network_only\": true},\"onboarding\": {\"onboarding_flow_pending\": true,\"signup_form_pending\": true}}"),
+			expectedStatus: http.StatusUnprocessableEntity,
+			expectedArray:  false,
+		},
 		{
 			name:           "PutAccount OK setting dashboard_features agent_network",
 			expectedBody:   true,

management/server/peer.go

@@ -106,11 +106,13 @@ func (am *DefaultAccountManager) MarkPeerConnected(ctx context.Context, peerPubK
 	}
 	if !updated {
 		am.metrics.AccountManagerMetrics().CountPeerStatusUpdate(telemetry.PeerStatusConnect, telemetry.PeerStatusStale)
-		log.WithContext(ctx).Tracef("peer %s already has a newer session in store, skipping connect", peer.ID)
+		log.WithContext(ctx).Debugf("peer %s already has a newer session in store, skipping connect", peer.ID)
 		return nil
 	}
 	am.metrics.AccountManagerMetrics().CountPeerStatusUpdate(telemetry.PeerStatusConnect, telemetry.PeerStatusApplied)
 
+	log.WithContext(ctx).Debugf("mark peer %s connected", peer.ID)
+
 	if err = am.schedulePeerExpirations(ctx, accountID, peer); err != nil {
 		return err
 	}
@@ -180,12 +182,14 @@ func (am *DefaultAccountManager) MarkPeerDisconnected(ctx context.Context, peerP
 	}
 	if !updated {
 		am.metrics.AccountManagerMetrics().CountPeerStatusUpdate(telemetry.PeerStatusDisconnect, telemetry.PeerStatusStale)
-		log.WithContext(ctx).Tracef("peer %s session token mismatch on disconnect (token=%d), skipping",
+		log.WithContext(ctx).Debugf("peer %s session token mismatch on disconnect (token=%d), skipping",
 			peer.ID, sessionStartedAt)
 		return nil
 	}
 	am.metrics.AccountManagerMetrics().CountPeerStatusUpdate(telemetry.PeerStatusDisconnect, telemetry.PeerStatusApplied)
 
+	log.WithContext(ctx).Debugf("mark peer %s disconnected", peer.ID)
+
 	// Symmetric with MarkPeerConnected: when an embedded proxy peer goes
 	// offline, refresh the peers that had synthesized records pointing at
 	// it so they pull the stale entries instead of waiting out TTL.

proxy/internal/middleware/builtin/llm_guardrail/middleware.go

@@ -25,6 +25,14 @@ const (
 	denyCodeModel    = "llm_policy.model_blocked"
 	denyReasonModel  = "model_blocked"
 	denyMessageModel = "model is not in the policy allowlist"
+	// Deny reason used when an allowlist is configured but the request model
+	// could not be determined. URL/path-routed providers (AWS Bedrock, Google
+	// Vertex, ...) carry the model outside the JSON body, so a request shape the
+	// parser does not recognise reaches the guardrail with no model. Such a
+	// request must be denied (fail closed), never waved through.
+	denyCodeModelUnknown    = "llm_policy.model_unknown"
+	denyReasonModelUnknown  = "model_unknown"
+	denyMessageModelUnknown = "request model could not be determined for the policy allowlist"
 )
 
 // Middleware enforces the model allowlist and optionally captures the
@@ -108,23 +116,37 @@ func (m *Middleware) evaluateAllowlist(model string, modelPresent bool) *middlew
 	if len(m.cfg.ModelAllowlist) == 0 {
 		return nil
 	}
-	if !modelPresent {
-		return nil
+	// Fail closed: with an allowlist configured, a request whose model the
+	// upstream parser could not extract (absent or empty) must be denied rather
+	// than allowed. This is what enforces the allowlist for URL/path-routed
+	// providers (Bedrock, Vertex, ...) whose model lives outside the JSON body.
+	if !modelPresent || normaliseModel(model) == "" {
+		return denyModel("", denyCodeModelUnknown, denyMessageModelUnknown, denyReasonModelUnknown)
 	}
 	if m.modelInAllowlist(model) {
 		return nil
 	}
+	return denyModel(model, denyCodeModel, denyMessageModel, denyReasonModel)
+}
+
+// denyModel builds a 403 deny Output for a model-allowlist rejection. model is
+// included in the details only when non-empty.
+func denyModel(model, code, message, reason string) *middleware.Output {
+	details := map[string]string{}
+	if model != "" {
+		details["model"] = model
+	}
 	return &middleware.Output{
 		Decision:   middleware.DecisionDeny,
 		DenyStatus: 403,
 		DenyReason: &middleware.DenyReason{
-			Code:    denyCodeModel,
-			Message: denyMessageModel,
-			Details: map[string]string{"model": model},
+			Code:    code,
+			Message: message,
+			Details: details,
 		},
 		Metadata: []middleware.KV{
 			{Key: middleware.KeyLLMPolicyDecision, Value: "deny"},
-			{Key: middleware.KeyLLMPolicyReason, Value: denyReasonModel},
+			{Key: middleware.KeyLLMPolicyReason, Value: reason},
 		},
 	}
 }

proxy/internal/middleware/builtin/llm_guardrail/middleware_test.go

@@ -102,13 +102,44 @@ func TestAllowlistCaseInsensitive(t *testing.T) {
 	}
 }
 
-func TestAllowlistMissingModelKeyAllows(t *testing.T) {
+func TestAllowlistMissingModelKeyDenies(t *testing.T) {
+	// Fail closed: with an allowlist configured, a request whose model the
+	// parser could not extract (URL/path-routed providers such as Bedrock or
+	// Vertex whose shape wasn't recognised) must be denied, not allowed.
 	mw := New(Config{ModelAllowlist: []string{"gpt-4o"}})
 	out, err := mw.Invoke(context.Background(), newInput())
 	require.NoError(t, err)
-	assert.Equal(t, middleware.DecisionAllow, out.Decision, "missing model key must allow even with non-empty allowlist")
+	require.NotNil(t, out)
+	assert.Equal(t, middleware.DecisionDeny, out.Decision, "absent model must be denied when an allowlist is set")
+	assert.Equal(t, 403, out.DenyStatus, "deny status must be 403")
+	require.NotNil(t, out.DenyReason, "deny reason must be populated")
+	assert.Equal(t, "llm_policy.model_unknown", out.DenyReason.Code, "deny code must be model_unknown")
 	dec, _ := metaValue(t, out.Metadata, middleware.KeyLLMPolicyDecision)
-	assert.Equal(t, "allow", dec, "decision must be allow when model key is absent")
+	assert.Equal(t, "deny", dec, "decision must be deny when model key is absent")
+	reason, _ := metaValue(t, out.Metadata, middleware.KeyLLMPolicyReason)
+	assert.Equal(t, "model_unknown", reason, "reason metadata must be model_unknown")
+}
+
+func TestAllowlistEmptyModelValueDenies(t *testing.T) {
+	// A present-but-empty model is as undeterminable as an absent one.
+	mw := New(Config{ModelAllowlist: []string{"gpt-4o"}})
+	out, err := mw.Invoke(context.Background(), newInput(
+		middleware.KV{Key: middleware.KeyLLMModel, Value: "   "},
+	))
+	require.NoError(t, err)
+	require.NotNil(t, out)
+	assert.Equal(t, middleware.DecisionDeny, out.Decision, "empty model must be denied when an allowlist is set")
+	require.NotNil(t, out.DenyReason, "deny reason must be populated")
+	assert.Equal(t, "llm_policy.model_unknown", out.DenyReason.Code, "deny code must be model_unknown")
+}
+
+func TestAllowlistEmptyListAllowsMissingModel(t *testing.T) {
+	// Without an allowlist there is nothing to enforce, so a missing model is
+	// still allowed — the fail-closed rule only applies when a list is set.
+	mw := New(Config{})
+	out, err := mw.Invoke(context.Background(), newInput())
+	require.NoError(t, err)
+	assert.Equal(t, middleware.DecisionAllow, out.Decision, "no allowlist must allow even without a model")
 }
 
 func TestPromptCaptureDisabledEmitsNoPrompt(t *testing.T) {

proxy/internal/middleware/builtin/llm_request_parser/guardrail_allowlist_test.go

@@ -0,0 +1,106 @@
+package llm_request_parser
+
+import (
+	"context"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/proxy/internal/middleware"
+	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin/llm_guardrail"
+)
+
+// runParserGuardrail runs the request parser then the model-allowlist guardrail
+// in SlotOnRequest order, threading the parser's metadata into the guardrail the
+// same way the real chain does. It returns the guardrail decision so tests can
+// assert allowlist enforcement for URL/path-routed providers end to end.
+func runParserGuardrail(t *testing.T, url string, body []byte, allowlist []string) *middleware.Output {
+	t.Helper()
+	parser := newMiddleware(t)
+	parsed, err := parser.Invoke(context.Background(), &middleware.Input{
+		Slot: middleware.SlotOnRequest,
+		URL:  url,
+		Body: body,
+	})
+	require.NoError(t, err, "parser must not error")
+
+	guard := llm_guardrail.New(llm_guardrail.Config{ModelAllowlist: allowlist})
+	out, err := guard.Invoke(context.Background(), &middleware.Input{
+		Slot:     middleware.SlotOnRequest,
+		Metadata: parsed.Metadata,
+	})
+	require.NoError(t, err, "guardrail must not error")
+	require.NotNil(t, out, "guardrail must return an output")
+	return out
+}
+
+// TestModelAllowlist_URLRoutedProviders validates that the model allowlist is
+// enforced for providers whose model travels in the URL path (AWS Bedrock,
+// Google Vertex) rather than the JSON body. The "unknown action" case is the
+// regression guard for #6751: a Bedrock request shape the parser cannot map to a
+// model must fail closed under an allowlist instead of bypassing it.
+func TestModelAllowlist_URLRoutedProviders(t *testing.T) {
+	const bedrockBody = `{"anthropic_version":"bedrock-2023-05-31","messages":[{"role":"user","content":"hi"}]}`
+	const vertexBody = `{"anthropic_version":"vertex-2023-10-16","messages":[{"role":"user","content":"hi"}]}`
+
+	tests := []struct {
+		name      string
+		url       string
+		body      string
+		allowlist []string
+		decision  middleware.Decision
+		denyCode  string
+	}{
+		{
+			name:      "bedrock allowed model passes",
+			url:       "https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-haiku-4-5-v1:0/invoke",
+			body:      bedrockBody,
+			allowlist: []string{"anthropic.claude-haiku-4-5"},
+			decision:  middleware.DecisionAllow,
+		},
+		{
+			name:      "bedrock disallowed model denied",
+			url:       "https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-opus-4-8-v1:0/invoke",
+			body:      bedrockBody,
+			allowlist: []string{"anthropic.claude-haiku-4-5"},
+			decision:  middleware.DecisionDeny,
+			denyCode:  "llm_policy.model_blocked",
+		},
+		{
+			name:      "bedrock unknown action fails closed",
+			url:       "https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-opus-4-8-v1:0/some-future-action",
+			body:      bedrockBody,
+			allowlist: []string{"anthropic.claude-haiku-4-5"},
+			decision:  middleware.DecisionDeny,
+			denyCode:  "llm_policy.model_unknown",
+		},
+		{
+			name:      "vertex disallowed model denied",
+			url:       "/v1/projects/p/locations/global/publishers/anthropic/models/claude-opus-4-8@20250101:rawPredict",
+			body:      vertexBody,
+			allowlist: []string{"claude-haiku-4-5"},
+			decision:  middleware.DecisionDeny,
+			denyCode:  "llm_policy.model_blocked",
+		},
+		{
+			name:      "vertex allowed model passes",
+			url:       "/v1/projects/p/locations/global/publishers/anthropic/models/claude-haiku-4-5@20250101:rawPredict",
+			body:      vertexBody,
+			allowlist: []string{"claude-haiku-4-5"},
+			decision:  middleware.DecisionAllow,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			out := runParserGuardrail(t, tt.url, []byte(tt.body), tt.allowlist)
+			assert.Equal(t, tt.decision, out.Decision, "unexpected decision for %s", tt.name)
+			if tt.decision == middleware.DecisionDeny {
+				require.NotNil(t, out.DenyReason, "deny reason must be set for %s", tt.name)
+				assert.Equal(t, 403, out.DenyStatus, "deny status must be 403 for %s", tt.name)
+				assert.Equal(t, tt.denyCode, out.DenyReason.Code, "deny code for %s", tt.name)
+			}
+		})
+	}
+}

shared/management/http/api/openapi.yml

@@ -376,7 +376,7 @@ components:
           type: boolean
           example: false
         agent_network_only:
-          description: Limits the dashboard to the Agent Network surface for this account. Set for accounts created via netbird.ai signups and can be disabled later.
+          description: Limits the dashboard to the Agent Network surface for this account. Set for accounts created via netbird.ai signups and can be disabled later. Enabling this requires dashboard_features.agent_network to be true in the same request.
           type: boolean
           example: false
         dashboard_features:
@@ -414,7 +414,7 @@ components:
       type: object
       properties:
         agent_network:
-          description: Controls the Agent Network menu for the account regardless of the deployment feature flag. When true the menu is shown, when false it is hidden, and when omitted the default behavior applies.
+          description: Controls the Agent Network menu for the account regardless of the deployment feature flag. When true the menu is shown, when false it is hidden, and when omitted the default behavior applies. Must be true when agent_network_only is enabled.
           type: boolean
           example: true
     AccountExtraSettings:

shared/management/http/api/types.gen.go

@@ -1614,7 +1614,7 @@ type Account struct {
 
 // AccountDashboardFeatures Per-account dashboard section visibility overrides. Omitted keys follow the default dashboard behavior.
 type AccountDashboardFeatures struct {
-	// AgentNetwork Controls the Agent Network menu for the account regardless of the deployment feature flag. When true the menu is shown, when false it is hidden, and when omitted the default behavior applies.
+	// AgentNetwork Controls the Agent Network menu for the account regardless of the deployment feature flag. When true the menu is shown, when false it is hidden, and when omitted the default behavior applies. Must be true when agent_network_only is enabled.
 	AgentNetwork *bool `json:"agent_network,omitempty"`
 }
 
@@ -1653,7 +1653,7 @@ type AccountRequest struct {
 
 // AccountSettings defines model for AccountSettings.
 type AccountSettings struct {
-	// AgentNetworkOnly Limits the dashboard to the Agent Network surface for this account. Set for accounts created via netbird.ai signups and can be disabled later.
+	// AgentNetworkOnly Limits the dashboard to the Agent Network surface for this account. Set for accounts created via netbird.ai signups and can be disabled later. Enabling this requires dashboard_features.agent_network to be true in the same request.
 	AgentNetworkOnly *bool `json:"agent_network_only,omitempty"`
 
 	// AutoUpdateAlways When true, updates are installed automatically in the background. When false, updates require user interaction from the UI.

shared/relay/client/early_msg_buffer.go

@@ -10,7 +10,7 @@ import (
 
 const (
 	earlyMsgTTL      = 5 * time.Second
-	earlyMsgCapacity = 1000
+	earlyMsgCapacity = 10000
 )
 
 // earlyMsgBuffer buffers transport messages that arrive before the corresponding