openclaw-mem0-memory

I was paying $20/month for Supermemory's OpenClaw memory plugin. It worked fine. Semantic search, auto-recall, auto-capture, user profiles. But $20/month for something that runs a vector DB and an LLM extraction step felt off.

So I forked their plugin and replaced the entire backend with Mem0 (Apache 2.0, self-hosted). Same hooks, same tools, same OpenClaw integration. Zero cost beyond the API calls you're already paying for.

How it works

The plugin registers hooks into OpenClaw's lifecycle:

• before_agent_start: queries your memory store, injects relevant context into the prompt
• agent_end: captures the conversation, sends it to Mem0 for fact extraction and storage

Mem0 handles the hard parts. It extracts facts from raw conversation text, deduplicates against existing memories, and maintains a vector store for semantic search.

The architecture is simple:

OpenClaw plugin (TypeScript)
    |
    v
Mem0 REST server (FastAPI, server/)
    |
    +-- ChromaDB (local vector store)
    +-- GPT-5-Mini or Claude (fact extraction)
    +-- text-embedding-3-small (embeddings)

I migrated 1141 memories from Supermemory using the included migration script. Semantic search works perfectly. The recall quality is the same or better since Mem0's extraction is solid.

Setup

1. Install the plugin

# copy this repo to your openclaw plugins directory
cp -r . ~/.openclaw/plugins/openclaw-mem0-memory
cd ~/.openclaw/plugins/openclaw-mem0-memory
bun install

2. Start the Mem0 server

pip install mem0ai fastapi uvicorn chromadb aiohttp
cd server
python server.py

The server runs on port 8080 by default. Set MEM0_PORT env var to change it.

You need either OPENAI_API_KEY (for GPT-5-Mini extraction + embeddings) or ANTHROPIC_API_KEY (for Claude extraction) plus OPENAI_API_KEY (embeddings always use OpenAI). The server loads env vars from ~/.openclaw/workspace/.env automatically.

3. Configure OpenClaw

Add to ~/.openclaw/openclaw.json:

{
  "plugins": {
    "entries": {
      "openclaw-mem0-memory": {
        "enabled": true,
        "config": {
          "mem0Url": "http://localhost:8080",
          "autoRecall": true,
          "autoCapture": true
        }
      }
    }
  }
}

Restart OpenClaw. Done.

Tools and Commands

The AI gets four tools automatically:

Tool	What it does
mem0_store	Save information to long-term memory
mem0_search	Semantic search across memories
mem0_forget	Delete a memory by query or ID
mem0_profile	View user profile (persistent facts + recent context)

Slash commands: /remember <text> and /recall <query>.

CLI: openclaw mem0 status, openclaw mem0 search <query>, openclaw mem0 profile, openclaw mem0 wipe.

Configuration

Key	Type	Default	Description
mem0Url	string	http://localhost:8080	Mem0 server URL
userId	string	openclaw_{hostname}	Memory namespace
autoRecall	boolean	true	Inject memories before every AI turn
autoCapture	boolean	true	Store conversations after every turn
maxRecallResults	number	10	Max memories injected per turn
profileFrequency	number	50	Full profile injection every N turns
captureMode	string	"all"	"all" filters noise, "everything" captures all
debug	boolean	false	Verbose logging

Migration from Supermemory

If you have existing memory files in ~/.openclaw/workspace/memory/, run:

cd server
python migrate.py

This reads all .md files, chunks them by section, and ingests everything into Mem0 concurrently. Edit USER_ID in migrate.py to match your config.

License

MIT