🦞🧠 OpenClaw EverMemOS Plugin

An OpenClaw plugin that adds long-term memory powered by EverMemOS — an advanced Memory OS that provides structured extraction, intelligent retrieval, and progressive profile building with 93% reasoning accuracy on the LoCoMo benchmark.

Overview

This plugin integrates EverMemOS as the memory layer for OpenClaw agents. Unlike simple key-value memory stores, EverMemOS provides:

• Multiple memory types — Episodic memories, event logs, foresight predictions, and user profiles
• Smart retrieval — BM25 keyword search, vector semantic search, hybrid, RRF fusion, and LLM-guided agentic retrieval
• Structured extraction — Automatically extracts structured memories from conversations with boundary detection
• Progressive profiles — Builds and evolves user profiles over time from conversation patterns

How It Works

                     ┌──────────────────┐
 OpenClaw Agent ────▶│  openclaw-       │──── REST API ──── EverMemOS Server
                     │  evermemos       │                  (localhost:1995)
                     │  plugin          │
                     └──────────────────┘
                       │
                       ├─ Auto-Recall (before_agent_start)
                       │   Searches relevant memories and injects into context
                       │
                       ├─ Auto-Capture (agent_end)
                       │   Stores conversation messages for memory extraction
                       │
                       └─ 5 Agent Tools
                           memory_search, memory_store, memory_get,
                           memory_list, memory_forget

1. Auto-Recall — Before the agent responds, the plugin searches EverMemOS for memories relevant to the current message and injects them into the agent's context
2. Auto-Capture — After the agent responds, the last user/assistant exchange is sent to EverMemOS which extracts episodic memories, event logs, foresight predictions, and updates user profiles
3. Agent Tools — Five tools for explicit memory operations during conversations
4. Slash Commands — /remember and /recall for quick manual memory operations

Prerequisites

• OpenClaw installed and configured
• EverMemOS server running (default: http://localhost:1995)

Setting up EverMemOS

# Clone EverMemOS
git clone https://github.com/EverMind-AI/EverMemOS.git
cd EverMemOS

# Start infrastructure services
docker compose up -d

# Install dependencies
uv sync

# Configure environment
cp env.template .env
# Edit .env with your LLM_API_KEY and VECTORIZE_API_KEY

# Start the server
uv run python src/run.py

# Verify it's running
curl http://localhost:1995/health

Installation

openclaw plugins install @zhenhangtung/openclaw-evermemos

Configuration

Add to your openclaw.json:

Minimal Configuration

// plugins.entries
"openclaw-evermemos": {
  "enabled": true,
  "config": {
    // Self-hosted: http://localhost:1995/api/v1
    // Cloud: https://api.evermind.ai/api/v0
    "baseUrl": "http://localhost:1995/api/v1",

    "userId": "your-user-id"
  }
}

Full Configuration

"openclaw-evermemos": {
  "enabled": true,
  "config": {
    // EverMemOS API base URL (must include versioned API path, supports ${EVERMEMOS_BASE_URL} env var)
    // Self-hosted: "http://localhost:1995/api/v1"
    // Cloud: "https://api.evermind.ai/api/v0"
    "baseUrl": "http://localhost:1995/api/v1",

    // API key: required for cloud, optional for self-hosted
    // Supports ${EVERMEMOS_API_KEY} env var
    "apiKey": "${EVERMEMOS_API_KEY}",

    // User ID for scoping memories
    "userId": "your-user-id",

    // Optional group ID for multi-user scenarios
    "groupId": "project-team",

    // Auto-recall: inject memories before each agent turn (default: true)
    "autoRecall": true,

    // Auto-capture: store context after each agent turn (default: true)
    "autoCapture": true,

    // Retrieval method: "keyword", "vector", "hybrid", "rrf", "agentic"
    "retrieveMethod": "hybrid",

    // Memory types to retrieve: "episodic_memory", "foresight", "event_log", "profile"
    "memoryTypes": ["episodic_memory"],

    // Maximum number of memories to retrieve per search (default: 10)
    "topK": 10
  }
}

Agent Tools

The agent gets five tools it can call during conversations:

| Tool | Description | |------|-------------| | memory_search | Search memories by natural language query with configurable retrieval method | | memory_store | Store a message for memory extraction (episodic, event log, foresight, profile) | | memory_get | Retrieve memories for a user by memory type | | memory_list | List all stored memories across all memory types | | memory_forget | Delete memories by event ID, user ID, or search query |

Memory Types

EverMemOS organizes memories into four types:

| Type | Description | |------|-------------| | episodic_memory | Narrative memories of events and conversations | | event_log | Atomic facts extracted from episodes | | foresight | Predictions and upcoming events extracted from conversations | | profile | Progressive user profiles built from conversation patterns |

Retrieval Methods

| Method | Description | |--------|-------------| | keyword | BM25 keyword search | | vector | Semantic vector search | | hybrid | Combined keyword + vector search | | rrf | Reciprocal Rank Fusion (keyword + vector + ranking fusion) | | agentic | LLM-guided multi-round intelligent retrieval |

Slash Commands

| Command | Description | |---------|-------------| | /remember <text> | Save something to EverMemOS memory | | /recall <query> | Search your memories and see results with relevance scores |

CLI Commands

# Search memories
openclaw evermemos search "what does the user like to do on weekends"

# Search with specific retrieval method
openclaw evermemos search "project deadlines" --method agentic

# View statistics
openclaw evermemos stats

Configuration Options

| Key | Type | Default | Description | |-----|------|---------|-------------| | baseUrl | string | http://localhost:1995/api/v1 | EverMemOS API URL. Must include /api/vx (e.g. /api/v1 or /api/v0) (supports ${EVERMEMOS_BASE_URL}) | | apiKey | string | — | API key for Bearer auth. Required for cloud, optional for self-hosted (supports ${EVERMEMOS_API_KEY}) | | userId | string | "default" | User ID for scoping memories | | groupId | string | — | Optional group ID for multi-user scenarios | | autoRecall | boolean | true | Inject memories before each agent turn | | autoCapture | boolean | true | Store context after each agent turn | | retrieveMethod | string | "hybrid" | Retrieval method: keyword, vector, hybrid, rrf, agentic | | memoryTypes | string[] | ["episodic_memory"] | Memory types to retrieve | | topK | number | 10 | Max memories per recall/search |

Development

# Install dependencies
npm install

# Run tests
npm test

# Type check
npm run lint

# Build
npm run build

Comparison with Mem0 Plugin

| Feature | openclaw-mem0 (Mem0) | openclaw-evermemos (EverMemOS) | |---------|---------------------|-------------------------------| | Memory types | Flat text memories | Episodic, event log, foresight, profile | | Retrieval | Similarity search | Keyword, vector, hybrid, RRF, agentic | | Profile building | Manual | Automatic progressive profiles | | Boundary detection | No | Automatic conversation boundary detection | | LoCoMo accuracy | — | 93% | | Backend | Mem0 Cloud or OSS | Self-hosted EverMemOS | | Foresight | No | Predictive/prospective memory |

License

MIT

Acknowledgments

• EverMemOS — The Memory OS for AI agents
• OpenClaw — Personal AI assistant
• Mem0 OpenClaw Plugin — Reference implementation for OpenClaw memory plugins