OpenClaw Integration

Add long-term memory to your OpenClaw agent. Zero LLM on write. Zero config.

Install


openclaw plugins install @unforget-ai/openclaw

Then install the embedded server (one-time):


pipx install unforget-embed

Restart the gateway:


openclaw gateway restart

That’s it. No API keys, no Docker, no database setup required.

How It Works

The plugin hooks into OpenClaw’s lifecycle:

Before each prompt — retrieves relevant memories and injects them into the agent’s context
After each response — stores the conversation turn as a memory
Forget/Remember — detects intent and handles it directly (e.g., “forget that I like pizza”)

All memory is stored in an embedded PostgreSQL database (via pgserver) that starts automatically. Data persists in ~/.unforget/data/.

Usage

Just chat normally. Memory is automatic:


You: My name is Alex and I like sushi
Agent: Got it — I'll remember that, Alex.

/new (new session)

You: What's my name and what food do I like?
Agent: Your name is Alex and you like sushi.

You: Forget that I like sushi
Agent: Forgotten.

/new

You: What food do I like?
Agent: You haven't told me any food preferences.

Memory Tools

The plugin registers three tools the agent can use directly:

Tool	Description
`memory_recall`	Search long-term memories
`memory_store`	Save a fact to memory
`memory_forget`	Delete memories by search query

To enable them, add to your agent config in openclaw.json:


{
  "agents": {
    "list": [
      {
        "id": "main",
        "tools": {
          "alsoAllow": [
            "memory_recall",
            "memory_store",
            "memory_forget"
          ]
        }
      }
    ]
  }
}

Configuration

The plugin works with zero config. Optional settings:


{
  "plugins": {
    "entries": {
      "@unforget-ai/openclaw": {
        "enabled": true,
        "config": {
          "orgId": "openclaw",
          "autoRetain": true,
          "autoRecall": true,
          "autoRecallTopK": 10,
          "debug": false
        }
      }
    }
  }
}

Option	Default	Description
`orgId`	`"openclaw"`	Organization ID for memory scoping
`agentId`	auto	Agent ID (defaults to OpenClaw agent ID)
`autoRetain`	`true`	Store conversation turns automatically
`autoRecall`	`true`	Inject relevant memories before each prompt
`autoRecallTopK`	`10`	Max memories to inject per prompt
`debug`	`false`	Enable debug logging

External Server

If you already have Unforget running or want to use your own PostgreSQL:


{
  "config": {
    "apiUrl": "http://localhost:9077"
  }
}

Architecture


OpenClaw agent
    │ hooks: before_agent_start, agent_end
    ▼
@unforget-ai/openclaw plugin (TypeScript)
    │ HTTP to localhost:9077
    ▼
unforget-embed (auto-started Python daemon)
    ├── unforget core (4-channel retrieval, embedder, reranker)
    └── pgserver (embedded PostgreSQL + pgvector)

Retrieval

Every memory recall runs 4 search channels in parallel:

Channel	What it does
Semantic	pgvector cosine similarity
BM25	PostgreSQL full-text search
Entity	Named entity overlap (people, places, dates)
Temporal	Recently accessed memories first

Results are fused with Reciprocal Rank Fusion and reranked with a cross-encoder. One SQL round trip.

Channels

The plugin works across all OpenClaw channels — Telegram, web UI, CLI, and any other connected channel. Memories are shared across channels for the same agent.

Troubleshooting

Plugin not loading?


tail -f ~/.openclaw/logs/gateway.log | grep -i unforget

Server not starting? Make sure unforget-embed is installed:


pipx install unforget-embed

Memories not recalling? Enable debug mode in openclaw.json:


{
  "config": {
    "debug": true
  }
}