MemMachine OpenClaw Plugin

This plugin integrates OpenClaw with MemMachine to provide persistent, queryable long-term memory across agent sessions. MemMachine (by MemVerge) stores interaction history and retrieves high-relevance context at inference time, enabling response grounding while reducing prompt size and token usage.

The plugin registers the following functions in OpenClaw:

• memory_search
• memory_store
• memory_forget
• memory_get

It also registers two CLI functions:

• search: Search MemMachine memory.
• stats: Retrieve stats from MemMachine.

Features

Auto Recall

When auto recall is enabled, the plugin searches episodic and semantic memories before the agent responds. Matching entries are injected into the context.

Auto Capture

When auto capture is enabled, the plugin sends each exchange to MemMachine after the agent responds.

Setup

Install from package registry

openclaw plugins install @memmachine/openclaw-memmachine

Install from local filesystem

openclaw plugins install ./MemMachine/integrations/openclaw

Install from a packed tarball

openclaw plugins install ./memmachine-openclaw-memmachine-0.0.0-development.tgz

Do not use openclaw hooks install for this package. It is an OpenClaw plugin pack that exports openclaw.extensions, not a hook pack that exports openclaw.hooks.

Platform (MemMachine Cloud)

Get an API key from MemMachine Cloud.

Configuration

You can configure the MemMachine plugin in the UI or by editing the memmachine entry in the openclaw.json file.

MemMachine configuration in openclaw.json

Here is a sample openclaw.json entry:

{
  "plugins": {
    "slots": {
      "memory": "openclaw-memmachine"
    },
    "entries": {
      "openclaw-memmachine": {
        "enabled": true,
        "config": {
          "apiKey": "mm-...",
          "baseUrl": "https://api.memmachine.ai",
          "autoCapture": true,
          "autoRecall": true,
          "orgId": "openclaw",
          "projectId": "openclaw",
          "searchThreshold": 0.5,
          "topK": 5,
          "userId": "openclaw"
        }
      }
    }
  }
}

Configuration entries

Here are the required configuration entries:

• apiKey: MemMachine API key.
• baseUrl: MemMachine API base URL.
• autoCapture: Enable automatic memory capture.
• autoRecall: Enable automatic memory recall.
• orgId: Organization identifier.
• projectId: Project identifier.
• searchThreshold: Minimum similarity score for recall.
• topK: Maximum number of memories to return.
• userId: User identifier for memory scoping.

Memory scoping and userId

The userId field controls how autoRecall scopes its search and how autoCapture tags stored memories. The behaviour depends on whether userId has been set to a real per-human-user value:

Default userId ("openclaw") — session-only recall

When userId is omitted or left as the default value "openclaw", the plugin uses the session's ephemeral sessionId as the sole discriminator. Recall is restricted to memories captured in the current conversation session (the session started by /new or /reset). Memories from earlier conversations are not retrieved.

This is the safe default. It prevents memories from one user's questions from leaking into another user's (or even the same user's later) questions.

Configured userId — cross-session long-term recall

When userId is set to a stable, per-human-user identifier — for example the Slack user ID of the person talking to the bot — autoRecall retrieves all memories previously stored under that user_id. This enables genuine long-term memory: preferences, decisions, and facts persist across separate sessions.

{
  "config": {
    "userId": "U012AB3CD"  // Slack user ID, Telegram user ID, etc.
  }
}

To use this correctly, set userId dynamically per user in your integration layer (e.g. per-channel account config, middleware, or environment variable). All users sharing the same userId value share their memory pool.

What happens when sessionId is unavailable

If OpenClaw does not supply a sessionId for a particular hook invocation (for example in some one-shot or legacy agent run modes), autoRecall skips the search entirely rather than issuing an unscoped query that would match all stored memories. You will see an info-level log line from the plugin when this occurs.