v1.0.5 final: tabby+flowise ripped out (verified broken), do-not-use recipes, GHCR token flow in update-check, 404-aware self-check, 10-service pack

README.md

@@ -132,22 +132,29 @@ basecamp wire           # wiring audit
 | # | Service | What it is | Port |
 |---|---------|-----------|------|
 | 1 | **code-server** | VS Code in the browser (the IDE) | 8443 |
-| 2 | **tabby** | Self-hosted AI code completion (Copilot replacement) | 8082 |
-| 3 | **qdrant** | Vector DB — code RAG backbone | 6333 |
-| 4 | **chroma** | Lightweight vector DB (alternative) | 8005 |
-| 5 | **n8n** | Workflow automation | 5678 |
-| 6 | **lobe-chat** | AI chat UI (wired to stack Ollama) | 3210 |
-| 7 | **anythingllm** | RAG workspace (docs + codebase Q&A) | 3001 |
-| 8 | **librechat** | Multi-model chat UI (needs mongo — included) | 3080 |
-| 9 | **meilisearch** | Full-text search (code/docs search) | 7700 |
-| 10 | **flowise** | No-code agent builder | 7860 |
-| 11 | **postgres** | Postgres + pgvector — Hermes long-term memory | 5432 |
-| 12 | **mongo** | MongoDB — LibreChat's database | 27017 |
-
-The IDE comes pre-loaded with **Continue.dev** (AI assistant) and the
-**Tabby** extension (inline completions). Point Continue at
-`http://ollama:11434/v1` (or `http://tabbyapi:5000/v1`) and Tabby at
-`http://starter-tabby:8082` — your stack's models power the whole pack.
+| 2 | **qdrant** | Vector DB — code RAG backbone | 6333 |
+| 3 | **chroma** | Lightweight vector DB (alternative) | 8005 |
+| 4 | **n8n** | Workflow automation | 5678 |
+| 5 | **lobe-chat** | AI chat UI (wired to stack Ollama) | 3210 |
+| 6 | **anythingllm** | RAG workspace (docs + codebase Q&A) | 3001 |
+| 7 | **librechat** | Multi-model chat UI (needs mongo — included) | 3080 |
+| 8 | **meilisearch** | Full-text search (code/docs search) | 7700 |
+| 9 | **postgres** | Postgres + pgvector — Hermes long-term memory | 5432 |
+| 10 | **mongo** | MongoDB — LibreChat's database | 27017 |
+
+**Removed (verified broken, 2026-08-09):** `tabby` (its bundled llama-server
+dlopens libcuda.so.1 even with `--device cpu` — crash-loops in CPU-only
+containers; needs GPU access) and `flowise` (upstream dependency bug —
+`@langchain/core` export error crashes boot on `latest` and 3.1.4).
+Both have full "do not use" recipes in basecamp so the agent warns before
+anyone ships them again.
+
+The IDE comes pre-loaded with **Continue.dev** (AI assistant). Point
+Continue's model at `http://ollama:11434/v1` (or
+`http://tabbyapi:5000/v1`) — your stack's models power the whole pack.
+(Inline code completion via the Tabby extension is available but needs a
+completion server; tabby's own image is currently broken in CPU-only
+containers — see the recipe.)
 
 Default credentials are in the compose file — change them before exposing
 anything publicly.

docker-compose.starter.yml

@@ -20,7 +20,6 @@ networks:
 
 volumes:
   code-server-data:
-  tabby-data:
   qdrant-data:
   chroma-data:
   n8n-data:
@@ -54,24 +53,15 @@ services:
       exec dumb-init /usr/bin/code-server --bind-addr 0.0.0.0:8080 --disable-telemetry
       "
 
-  # ── 2. AI CODE COMPLETION — self-hosted Copilot ──
-  tabby:
-    image: tabbyml/tabby:20260330
-    container_name: starter-tabby
-    restart: unless-stopped
-    networks: [tabby-tavern_ai-network]
-    ports:
-      - "8082:8080"              # completion server UI/API
-    volumes:
-      - tabby-data:/data
-    environment:
-      - TABBY_MODEL_CACHE_DIR=/data/models
-    command: serve --device cpu --model StarCoder2-3B --chat-model StarCoder2-3B
-    # --device cpu: keep completions off the 12GB card (TabbyAPI owns it).
-    # Version 20260330's serve has NO embedding supervisor (the crashy
-    # Nomic-Embed subprocess with -ngl 9999 only exists in newer 'latest').
-    # NOTE: for GPU-backed completion, use --device cuda and a larger model;
-    # the stack's TabbyAPI (tabbyapi:5000) serves the chat side.
+  # ── 2. (removed) tabby — self-hosted Copilot
+  # REMOVED 2026-08-09: tabby's bundled llama-server dlopens libcuda.so.1 at
+  # startup even with --device cpu (binary is CUDA-compiled), and the
+  # container has no NVIDIA libs. Verified: 'error while loading shared
+  # libraries: libcuda.so.1' → exit 127 crash-loop on 20260330 pin too.
+  # Would need --gpus all + CUDA libs — the 12GB card belongs to TabbyAPI.
+  # Completions alternative: point the code-server Tabby extension at a
+  # completion-capable model on the stack's own engines.
+  # Re-add when the tabby image ships a true CPU-only build.
 
   # ── 3. VECTOR DB — code RAG backbone ──
   qdrant:
@@ -152,6 +142,7 @@ services:
       - librechat-data:/app/librechat
     environment:
       - MONGO_URI=mongodb://mongo:27017/LibreChat
+      - JWT_SECRET=basecamp-jwt-secret-change-me   # required (JwtStrategy)
       - OPENAI_API_KEY=ollama
       - OPENAI_API_BASE=http://ollama:11434/v1    # wired to stack Ollama
       - ALLOW_SOCIAL_LOGIN=true
@@ -181,20 +172,13 @@ services:
       - MEILI_MASTER_KEY=basecamp-dev-key         # change me
       - MEILI_ENV=development
 
-  # ── 10. NO-CODE AGENT BUILDER ──
-  flowise:
-    image: flowiseai/flowise:3.1.4
-    container_name: starter-flowise
-    restart: unless-stopped
-    networks: [tabby-tavern_ai-network]
-    ports:
-      - "7860:3000"
-    volumes:
-      - flowise-data:/root/.flowise
-    environment:
-      - PORT=3000
-      - FLOWISE_USERNAME=admin
-      - FLOWISE_PASSWORD=basecamp                 # change me
+  # ── 10. (removed) flowise — no-code agent builder
+  # REMOVED 2026-08-09: flowiseai/flowise is BROKEN upstream (both 'latest'
+  # and 3.1.4 crash at boot with ERR_PACKAGE_PATH_NOT_EXPORTED —
+  # @langchain/langgraph-checkpoint requires './utils/uuid' which
+  # @langchain/core doesn't export). Verified on both tags. Ripped out
+  # rather than shipping a known-broken tool. Re-add when upstream fixes
+  # their dependency tree.
 
   # ── 11. LONG-TERM MEMORY — Postgres + pgvector ──
   # Hermes's pluggable memory backend: vector store for embeddings so the

recipes.py

@@ -271,9 +271,9 @@ WIRING_RECIPES = {
     "tabby": {
         "where": "CLI flags at container start + /data volume for models.",
         "keys": "command: serve --device cpu|--device cuda --model <name> --chat-model <name>; env: TABBY_MODEL_CACHE_DIR; API: /v1/health, /v1/completions (code completion)",
-        "fix": "For CPU: `serve --device cpu --model StarCoder2-3B --chat-model StarCoder2-3B`. The VS Code extension points at http://<host>:8082. Completions are separate from chat — TabbyAPI serves chat, Tabby serves inline code completion.",
-        "verify": "curl http://<host>:8082/v1/health returns {\"health\":\"ok\"}.",
-        "pitfalls": "KNOWN CRASH (verified 2026-08): newer 'latest' images spawn a Nomic-Embed subprocess with `-ngl 9999` (all layers to GPU) that exits 127 on constrained GPUs (12GB shared with other inference), crash-looping the container. FIX: pin tabbyml/tabby:20260330 — its `serve` has NO embedding supervisor, so nothing crashes; run `--device cpu` to keep completions off the card. NOTE: `--disable-embedding` is NOT a valid flag in 20260330 (rejected at CLI parse). The embedding/chat models download on first boot — first start is slow.",
+        "fix": "DO NOT USE in CPU-only containers — BROKEN as of 2026-08-09 (verified on 20260330 pin): tabby's bundled llama-server dlopens libcuda.so.1 at startup EVEN with --device cpu (binary is CUDA-compiled), and crashes with exit 127 'cannot open shared object file' when the container lacks NVIDIA libs. Needs --gpus all + CUDA libs. Removed from the starter pack; re-add only with GPU access or a true CPU-only tabby build.",
+        "verify": "curl http://<host>:8082/v1/health — expected to FAIL in CPU-only containers until tabby ships a CPU-only build.",
+        "pitfalls": "VERIFIED 2026-08-09: (1) 'latest' images spawn Nomic-Embed with -ngl 9999 → exit 127 on constrained GPUs; (2) even the 20260330 pin crash-loops in CPU-only containers: llama-server dlopens libcuda.so.1 → 'cannot open shared object file' → exit 127. (3) StarCoder2-3B is completion-only (no chat template) — a separate instruct model is required for --chat-model. Ripped out of the pack.",
     },
     "meilisearch": {
         "where": "Env vars.",
@@ -291,9 +291,9 @@ WIRING_RECIPES = {
     "flowise": {
         "where": "Env vars (PORT, FLOWISE_USERNAME, FLOWISE_PASSWORD) + /root/.flowise volume.",
         "keys": "PORT=3000, FLOWISE_USERNAME, FLOWISE_PASSWORD; API: /api/v1/ping",
-        "fix": "Set FLOWISE_USERNAME/PASSWORD for auth. Pin a VERSIONED tag — `flowiseai/flowise:latest` is BROKEN as of 2026-08 (ERR_PACKAGE_PATH_NOT_EXPORTED in @langchain/core ./utils/uuid crashes node on startup). Use flowiseai/flowise:3.1.4 or newer.",
-        "verify": "curl http://<host>:7860/api/v1/ping returns pong.",
-        "pitfalls": "KNOWN BUG (verified 2026-08): latest image crashes at boot in ReActAgentChat/ReActAgentLLM nodes with '@langchain/core' exports error. Pin 3.1.4+. Login with FLOWISE_USERNAME/FLOWISE_PASSWORD.",
+        "fix": "DO NOT USE — BROKEN UPSTREAM as of 2026-08-09 (verified on latest AND 3.1.4): crashes at boot with ERR_PACKAGE_PATH_NOT_EXPORTED — @langchain/langgraph-checkpoint requires './utils/uuid' which @langchain/core doesn't export. Removed from the starter pack. Re-add only when upstream fixes their dependency tree.",
+        "verify": "curl http://<host>:7860/api/v1/ping — expected to FAIL until upstream fixes the langchain dep.",
+        "pitfalls": "KNOWN BUG (verified 2026-08-09, both latest and 3.1.4): ERR_PACKAGE_PATH_NOT_EXPORTED './utils/uuid' in @langchain/core crashes node at boot (ReActAgentChat/ReActAgentLLM nodes). Not a compose problem — the image's dependency tree is broken. Ripped out of the pack.",
     },
     "postgres": {
         "where": "Container env (db/user/password) + data volume.",