ClawMemory Skill

Sovereign agent memory engine — self-hosted, privacy-first. All data stays local (SQLite) with optional Turso cloud sync.

Repo: https://github.com/clawinfra/clawmemory
Server port: localhost:7437
Last verified: 2026-03-28

✅ VERIFIED WORKING PATTERNS (copy-paste ready)

Start the server

cd /tmp/clawmemory && ./clawmemory serve --config config.json
# OR with defaults (SQLite at ./clawmemory.db, port 7437, Ollama at localhost:11434)
./clawmemory serve

Guard rules:

• Ollama must be running for vector search — if not, BM25-only mode activates automatically (no crash)
• Server binds localhost:7437 by default — not exposed externally
• First run auto-runs migrations (safe to restart)

Store a fact manually

curl -s -X POST http://localhost:7437/facts \
  -H "Content-Type: application/json" \
  -d '{"text": "User prefers Python over Go for scripting", "category": "preference", "importance": 0.8}'

Search memory

# Hybrid BM25 + vector (best quality)
curl -s "http://localhost:7437/search?q=python+preference&limit=5" | python3 -m json.tool

# BM25-only (fast, no Ollama needed)
curl -s "http://localhost:7437/search?q=python+preference&limit=5&mode=bm25" | python3 -m json.tool

Extract facts from a conversation turn (auto-capture)

curl -s -X POST http://localhost:7437/extract \
  -H "Content-Type: application/json" \
  -d '{
    "turns": [
      {"role": "user", "content": "I always deploy to Hetzner, never AWS."},
      {"role": "assistant", "content": "Got it, using Hetzner for deployments."}
    ]
  }' | python3 -m json.tool

Get user profile

curl -s http://localhost:7437/profile | python3 -m json.tool

Forget a fact

curl -s -X DELETE http://localhost:7437/facts/<fact-id>

OpenClaw Plugin (TypeScript) — Auto-wire

The plugin at plugin/ auto-injects memory pre-turn and auto-captures post-turn.

cd /tmp/clawmemory/plugin && npm install && npm run build
# Copy plugin/dist/ to OpenClaw plugins dir and enable in config

Plugin config in openclaw.config.json:

{
  "plugins": [
    {
      "id": "clawmemory",
      "path": "./plugins/clawmemory/dist/index.js",
      "config": {
        "serverUrl": "http://localhost:7437",
        "maxContextFacts": 10,
        "minImportance": 0.3
      }
    }
  ]
}

What it does automatically:

• Pre-turn: searches memory for relevant facts → injects as [Memory context] block into system prompt
• Post-turn: sends conversation turn to /extract → stores new facts

Config Reference (config.json)