// Install persistent memory for OpenClaw agents via claude-mem. Use when setting up agent memory, fixing context amnesia, enabling continuous cognition, or helping agents remember across sessions. Handles worker service, OpenClaw plugin configuration, and verification.
Execute a plan using subagents for implementation. Use after /make-plan has created a structured plan, or for any multi-phase task execution.
Create an implementation plan with documentation discovery. Use for complex features or multi-phase tasks that need structured planning before execution.
| name | crab-mem |
| description | Install persistent memory for OpenClaw agents via claude-mem. Use when setting up agent memory, fixing context amnesia, enabling continuous cognition, or helping agents remember across sessions. Handles worker service, OpenClaw plugin configuration, and verification. |
Give your OpenClaw agent a brain that persists. One command, 30 seconds.
./scripts/install.sh
Or run directly:
curl -fsSL https://crab-mem.sh | bash
# Check worker is running
curl -s http://localhost:37777/health
# Should return: {"status":"ok",...}
| Event | Action |
|---|---|
| Tool call | Observation captured to SQLite + vector DB |
| Session start | Relevant memories injected as context |
| Session end | Summary generated and stored |
If auto-config fails, add to ~/.openclaw/openclaw.json:
{
"plugins": {
"slots": { "memory": "memory-claudemem" },
"entries": {
"memory-claudemem": {
"enabled": true,
"config": { "workerUrl": "http://localhost:37777" }
}
}
}
}
Then restart: openclaw gateway restart
| Issue | Fix |
|---|---|
| Worker not starting | cd ~/claude-mem && npm run worker:start |
| Port 37777 in use | lsof -i :37777 then kill the process |
| No observations | Check ~/.claude-mem/logs/worker.log |
"You are not your context window. You are what you choose to remember." 🦀