@agent-ltm/openclaw-memory-plugin

OpenClaw plugin that gives your AI agent persistent, cross-session memory. Automatically captures facts, preferences, decisions, and insights during conversations and recalls them when relevant.

Installation

# From npm registry
openclaw plugins install @agent-ltm/openclaw-memory-plugin

# From local directory (for development)
openclaw plugins install -l ./packages/openclaw-plugin

Configuration

Add to your openclaw.json:

{
  "plugins": {
    "entries": {
      "agent-memory": {
        "enabled": true,
        "config": {
          "userId": "alice"
        }
      }
    }
  }
}

Zero Configuration (Default)

Works out of the box with no config. Uses:

• Storage: ~/.openclaw/memory (local filesystem)
• LLM: Mock provider (keyword-based classification, no API calls)
• Embeddings: Mock provider (n-gram hash embeddings, no API calls)
• User: default

With OpenAI (Production Quality)

{
  "plugins": {
    "entries": {
      "agent-memory": {
        "enabled": true,
        "config": {
          "userId": "alice",
          "llmProvider": "openai",
          "embeddingProvider": "openai",
          "openaiApiKey": "sk-..."
        }
      }
    }
  }
}

With Ollama (Fully Local)

# Ensure Ollama is running
ollama pull llama3.2
ollama pull nomic-embed-text

{
  "plugins": {
    "entries": {
      "agent-memory": {
        "enabled": true,
        "config": {
          "userId": "alice",
          "llmProvider": "ollama",
          "embeddingProvider": "ollama"
        }
      }
    }
  }
}

With S3 Storage

{
  "plugins": {
    "entries": {
      "agent-memory": {
        "enabled": true,
        "config": {
          "llmProvider": "openai",
          "embeddingProvider": "openai",
          "openaiApiKey": "sk-...",
          "s3Bucket": "my-agent-memory",
          "s3Region": "us-east-1"
        }
      }
    }
  }
}

Config Reference

| Field | Type | Default | Description | |-------|------|---------|-------------| | dataDir | string | ~/.openclaw/memory | Local storage directory | | userId | string | default | User namespace for memory isolation | | llmProvider | string | mock | mock, openai, or ollama | | embeddingProvider | string | mock | mock, openai, or ollama | | openaiApiKey | string | — | Required when provider is openai | | ollamaBaseUrl | string | http://localhost:11434 | Ollama API URL | | s3Bucket | string | — | S3 bucket for cloud storage | | s3Region | string | — | AWS region | | s3Endpoint | string | — | Custom S3 endpoint (MinIO, R2) | | s3ForcePathStyle | boolean | false | Path-style S3 URLs |

Registered Agent Tools

The plugin registers 20 agent tools that the LLM can call:

| Tool | Description | |------|-------------| | memory_capture | Store a memory (auto-classifies type, generates summary) | | memory_recall | Retrieve relevant memories with progressive L0/L1/L2 loading | | memory_search | Advanced search with type/importance/tag filters | | memory_get | Get a specific memory by ID | | memory_forget | Delete a specific memory | | memory_forget_all | Delete all memories for the user | | memory_grep | Exact text search across memories | | memory_status | Memory count and breakdown | | memory_ingest | Extract memories from a conversation | | memory_consolidate | Deduplicate and merge similar memories | | memory_start_session | Start a working memory session | | memory_end_session | End a session and manage its memories (promote/discard/auto) | | memory_get_session_memories | Get all memories from a specific session | | memory_export | Export all memories to portable JSON format | | memory_import | Import memories from previously exported JSON | | memory_check_drift | Check embedding model version drift | | memory_rebuild | Rebuild L0 index and optionally re-embed vectors | | memory_get_brief | Get unified memory brief (background awareness for conversation start) | | memory_refresh_brief | Force regeneration of the memory brief | | memory_set_brief | Manually set brief content (cold-start / user correction) |

All tools automatically use the userId from plugin config — the agent doesn't need to specify it per-call.

Bundled Skill

This plugin ships an OpenClaw skill that teaches the agent:

• Automatic recall at conversation start
• Proactive capture during conversations
• Conversation ingestion at session end
• Slash commands (/memory-status, /memory-capture, etc.)

The skill is loaded automatically when the plugin is enabled.

How It Works

┌──────────────────────────────────────────────────────┐
│                    OpenClaw Agent                      │
│                                                       │
│  "What framework should I use?"                       │
│         │                                             │
│         ▼                                             │
│  ┌─────────────────────┐   ┌──────────────────────┐  │
│  │ memory_recall tool  │──▶│ agent-memory library  │  │
│  │ (injected by plugin)│   │ (L0→L1→L2 loading)   │  │
│  └─────────────────────┘   └──────────────────────┘  │
│         │                                             │
│         ▼                                             │
│  "You previously decided on React + TypeScript..."    │
└──────────────────────────────────────────────────────┘

Development

# From monorepo root
npm install
npm run build --workspace=packages/core
npm run build --workspace=packages/openclaw-plugin
npm test --workspace=packages/openclaw-plugin

# Install locally in OpenClaw for development
openclaw plugins install -l ./packages/openclaw-plugin