OpenClaw Plugin -- Mnemory

An OpenClaw plugin that provides long-term semantic memory backed by a mnemory server.

Features

• Auto-Recall: Automatically fetches relevant memories and injects them into the system prompt before each agent turn
• Auto-Capture: Extracts and stores memories from conversations after each exchange (fire-and-forget)
• 15 Explicit Tools: memory_search, memory_find, memory_ask, memory_add, memory_add_batch, memory_update, memory_delete, memory_delete_batch, memory_list, memory_categories, memory_recent, memory_save_artifact, memory_get_artifact, memory_list_artifacts, memory_delete_artifact
• CLI Commands: openclaw mnemory status, openclaw mnemory search, openclaw mnemory list
• Compaction-safe: Memories survive session compaction via hook-based re-injection

Prerequisites

A running mnemory server accessible via HTTP.

Install

openclaw plugins install @fpytloun/openclaw-mnemory

Configure

In your openclaw.json:

{
  plugins: {
    slots: { memory: "mnemory" },
    entries: {
      mnemory: {
        enabled: true,
        config: {
          url: "http://localhost:8050",
          apiKey: "${MNEMORY_API_KEY}"
        }
      }
    }
  }
}

Configuration Options

| Key | Env Var | Default | Description | |-----|---------|---------|-------------| | url | MNEMORY_URL | (required) | Mnemory server URL | | apiKey | MNEMORY_API_KEY | (empty) | Bearer token for authentication | | userId | MNEMORY_USER_ID | (empty) | User ID (required when API key is wildcard) | | agentPrefix | MNEMORY_AGENT_PREFIX | openclaw | Prefix for agent IDs (e.g., openclaw:main) | | autoRecall | -- | true | Auto-inject memories into context | | autoCapture | -- | true | Auto-extract memories from conversations | | recallFindFirst | -- | true | Use AI-powered search on first turn (higher quality, slower) | | recallSearchMode | -- | search | Search mode for subsequent turns: find or search | | scoreThreshold | -- | 0.5 | Min relevance score for recalled memories (0.0-1.0) | | includeAssistant | -- | true | Send assistant messages for extraction | | managed | -- | true | Include behavioral instructions in system prompt | | timeout | MNEMORY_TIMEOUT | 60000 | HTTP request timeout in ms |

Config values support ${ENV_VAR} syntax for environment variable resolution.

How It Works

1. session_start: Pre-fetches instructions and core memories from /api/recall (non-blocking)
2. before_prompt_build: Two-phase recall per turn:
   • Awaits the init recall (instructions + core memories) if still pending
   • Sends the current user prompt as a search query to /api/recall for topical memories
   • First turn uses AI-powered find mode (configurable via recallFindFirst), subsequent turns use fast search mode (configurable via recallSearchMode)
   • Search results are replaced each turn (not accumulated) — the server deduplicates via session tracking
   • Injects instructions via prependSystemContext (cacheable), core memories + search results via appendSystemContext
3. agent_end: Extracts the last user+assistant exchange, strips inbound metadata, sends to /api/remember for background extraction
4. after_compaction: Re-fetches memories and resets turn tracking after context compaction
5. session_end: Cleans up session state

Memory Slot

This plugin declares kind: "memory" and is activated by setting plugins.slots.memory: "mnemory". Only one memory plugin can be active at a time.

License

MIT