update-1.6.2

deployments/README.md

@@ -1,29 +1,16 @@
-# Deployment Kits — GLPI NeuroForge Mega v1.6.2
+# Distributed Deployment Kits (v1.6.2)
 
-This repository intentionally supports both the complete NeuroForge platform and stripped standalone operation.
+This directory contains three independent deployment kits:
 
-## Complete / distributed platform
+- `master/` - authoritative NeuroForge Master plus Agent, Knowledge, Control, optional SearXNG and optional Prometheus/Grafana.
+- `cpu-subagent/` - CPU worker for `vector.relink` / graph convergence.
+- `gpu-subagent/` - GPU worker plus local Ollama for `model.chat` and `model.embed`.
 
-- `master/` — authoritative NeuroForge Master plus GLPI Agent, Knowledge, Control, optional SearXNG, Prometheus and Grafana. Remote CPU/GPU workers connect to this node.
-- `cpu-subagent/` — disposable CPU worker with `cpu,vector.relink` capabilities.
-- `gpu-subagent/` — disposable GPU worker plus Ollama with `gpu,model.chat,model.embed` capabilities.
+The generated `NEUROFORGE_WORKER_TOKEN` is identical in all three `.env` files. Replace the RFC 5737 example IP addresses (`192.0.2.x`) with real reachable addresses before starting.
 
-The Master is the only authoritative owner of NeuroForge state. CPU/GPU workers use leases, heartbeats and fenced job completion.
+Recommended order:
+1. GPU subagent: `docker compose --profile monitoring up -d`
+2. CPU subagent: `docker compose --profile monitoring up -d`
+3. Master: fill GLPI credentials, then `docker compose --profile research --profile monitoring up -d`
 
-## Standalone core operation
-
-- `agent/` — GLPI Agent with local Knowledge vector backend; no NeuroForge, Control or Research dependency.
-- `knowledge/` — standalone Knowledge editor/service.
-- `ollama/` — standalone Ollama runtime.
-- `combined/` — Agent + Knowledge + Ollama on one host, still without NeuroForge.
-
-The standalone Agent and Knowledge kits share `runtime/knowledge` by default. The Agent mounts it read-only; Knowledge mounts it read-write.
-
-## Release rules
-
-- Project images are pinned by `IMAGE_TAG=1.6.2`; production compose files do not require `latest`.
-- Replace every `CHANGE_ME_...` placeholder before startup.
-- Never use `docker compose down -v` during an in-place upgrade unless loss of persistent state is intended.
-- For a v1.6.0/v1.6.1 NeuroForge data volume, keep the volume: v1.6.2 includes the v1.6.1 recovery/OOM hotfixes and startup compaction path.
-
-See each role's README/preflight and the root `README.md` for startup order.
+Only the Master holds authoritative NeuroForge state. Workers are disposable execution nodes.

deployments/agent/docker-compose.yml

@@ -24,6 +24,7 @@ services:
       DATA_DIR: /app/data
       KNOWLEDGE_DIR: /app/knowledge
       KNOWLEDGE_VECTOR_BACKEND: local
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       OUTCOME_LEARNING_ENABLED: "false"
       OUTCOME_RETRIEVAL_ENABLED: "false"
       BRAIN_ACTIVITY_URL: ""

deployments/combined/docker-compose.yml

@@ -33,6 +33,7 @@ services:
       DATA_DIR: /app/data
       KNOWLEDGE_DIR: /app/knowledge
       OLLAMA_URL: http://ollama:11434
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       OLLAMA_URLS: http://ollama:11434
       KNOWLEDGE_VECTOR_BACKEND: local
       OUTCOME_LEARNING_ENABLED: "false"

deployments/gpu-subagent/docker-compose.yml

@@ -39,6 +39,8 @@ services:
       NEUROFORGE_WORKER_MAX_CONCURRENCY: ${NEUROFORGE_GPU_WORKER_CONCURRENCY:-1}
       NEUROFORGE_WORKER_HEARTBEAT_INTERVAL: ${NEUROFORGE_WORKER_HEARTBEAT_INTERVAL:-15s}
       NEUROFORGE_WORKER_OLLAMA_URL: http://ollama:11434
+      NEUROFORGE_WORKER_OLLAMA_API_KEY: ${NEUROFORGE_WORKER_OLLAMA_API_KEY:-}
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       NEUROFORGE_WORKER_OLLAMA_CHAT_MODEL: ${OLLAMA_MODEL:-gemma4}
       NEUROFORGE_WORKER_OLLAMA_EMBEDDING_MODEL: ${OLLAMA_EMBEDDING_MODEL:-embeddinggemma}
       NEUROFORGE_WORKER_OLLAMA_NUM_CTX: ${NEUROFORGE_OLLAMA_NUM_CTX:-8192}

deployments/knowledge/docker-compose.yml

@@ -11,6 +11,7 @@ services:
       BACKUP_DIR: /data/backups
       STAGING_DIR: /data/staging
       LISTEN_ADDR: :8080
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       BRAIN_ACTIVITY_URL: ""
       BRAIN_ACTIVITY_API_KEY: ""
     ports:

deployments/master/docker-compose.yml

@@ -47,6 +47,8 @@ services:
       OPENAI_API_KEY: ${OPENAI_API_KEY:-}
       NEUROFORGE_OLLAMA_URL: ${OLLAMA_BASE_URL:?Set Ollama URL reachable from master,
         normally GPU subagent host}
+      NEUROFORGE_OLLAMA_API_KEY: ${NEUROFORGE_OLLAMA_API_KEY:-}
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       NEUROFORGE_OLLAMA_CHAT_MODEL: ${OLLAMA_MODEL:-gemma3}
       NEUROFORGE_OLLAMA_EMBEDDING_MODEL: ${OLLAMA_EMBEDDING_MODEL:-embeddinggemma}
       NEUROFORGE_OLLAMA_NUM_CTX: ${NEUROFORGE_OLLAMA_NUM_CTX:-8192}
@@ -317,6 +319,7 @@ services:
       OLLAMA_THINK: ${OLLAMA_THINK:-}
       OLLAMA_TIMEOUT: ${OLLAMA_TIMEOUT:-}
       OLLAMA_URL: ${OLLAMA_BASE_URL:?Set Ollama URL reachable from master}
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       OLLAMA_URLS: ${OLLAMA_URLS:?Set one or more Ollama URLs reachable from master}
       OUTCOME_LEARNING_ENABLED: ${OUTCOME_LEARNING_ENABLED:-true}
       OUTCOME_LEARNING_FAIL_OPEN: ${OUTCOME_LEARNING_FAIL_OPEN:-false}
@@ -389,6 +392,7 @@ services:
       BASIC_AUTH_PASSWORD: ${BASIC_AUTH_PASSWORD:?Set the Knowledge web password}
       AI_FALLBACK_ENABLED: ${AI_FALLBACK_ENABLED:-false}
       OLLAMA_BASE_URL: ${OLLAMA_BASE_URL:?Set Ollama URL reachable from master}
+      OLLAMA_API_KEY: ${OLLAMA_API_KEY:-}
       OLLAMA_MODEL: ${OLLAMA_MODEL:-gemma3}
       OLLAMA_TIMEOUT: ${OLLAMA_TIMEOUT:-10m}
       OLLAMA_MAX_CONCURRENT: ${OLLAMA_MAX_CONCURRENT:-2}

deployments/master/monitoring/prometheus/alerts.yml

@@ -63,7 +63,7 @@ groups:
         labels: {severity: warning}
         annotations:
           summary: "NeuroForge pending job payload memory is high"
-          description: "Pending durable job payload/result bytes exceed 100 MiB for 5 minutes; backfill should remain below the v1.6.2 safety budget."
+          description: "Pending durable job payload/result bytes exceed 100 MiB for 5 minutes; backfill should remain below the v1.6.1 safety budget."
 
       - alert: NeuroForgeTerminalPayloadHigh
         expr: neuroforge_job_payload_bytes{state="terminal"} > 67108864
@@ -71,4 +71,4 @@ groups:
         labels: {severity: warning}
         annotations:
           summary: "NeuroForge terminal job payload retention is high"
-          description: "Terminal job payload/result bytes exceed 64 MiB. Completed vector.relink jobs should be compacted automatically in v1.6.2."
+          description: "Terminal job payload/result bytes exceed 64 MiB. Completed vector.relink jobs should be compacted automatically in v1.6.1."