360 Secure SkillHub
360 安全技能中心

Mem0 Config

面向 OpenClaw 的 mem0 长期记忆插件使用指南:讲解平台模式与开源模式的安装、Ollama 与 Qdrant 依赖、配置项(autoRecall、autoCapture、historyDbPath 等)、openclaw mem0 CLI 与 5 个记忆工具,并给出崩溃循环、服务连通性等排障与验证步骤。

nyrosveil
nyrosveil
热度 70.21
最近更新 2026-05-11 17:27:47
分类:知识依赖:代码执行 / 读写文件 / 第三方 API应用场景:笔记整理 / 工作流 / DevOps

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add mem0-config
Metadata
name
openclaw-mem0
description
Install, configure, diagnose, and operate the openclaw-mem0 long-term memory plugin for OpenClaw agents. Use when the user wants to set up mem0 memory (platform or open-source/self-hosted mode), configure OSS components (embedder, vector store, LLM, historyDbPath), debug memory issues (SQLITE_CANTOPEN crash loops, memories not being stored or recalled, Qdrant/Ollama connectivity), use the `openclaw mem0` CLI (search, stats), or understand how auto-recall and auto-capture work in agent sessions.

openclaw-mem0

Long-term memory plugin for OpenClaw agents, powered by Mem0. Extracts and injects memories automatically around each agent turn.

Installation

1. Install the plugin

openclaw plugins install @mem0/openclaw-mem0

This installs the plugin into ~/.openclaw/extensions/openclaw-mem0/ and adds it to openclaw.json.

2. Choose a mode and install dependencies

Platform mode — no local dependencies. Get an API key from app.mem0.ai and skip to Configuration.

Open-source mode — requires two local services:

# Ollama (embedder + LLM)
brew install ollama
ollama serve                        # or: brew services start ollama
ollama pull bge-m3:latest           # embedder (1024-dim)
ollama pull llama3.2                # LLM for memory extraction

# Qdrant (vector store)
docker run -d -p 6333:6333 qdrant/qdrant
# or: brew install qdrant && qdrant

Verify both are running:

curl -s http://localhost:11434/          # → "Ollama is running"
curl -s http://localhost:6333/health     # → {"status":"ok"}

3. Add to openclaw.json

Add under plugins.entries (see Configuration below).

4. Restart the gateway

openclaw gateway stop && openclaw gateway

Confirm the plugin loaded:

grep "openclaw-mem0: initialized" ~/.openclaw/logs/gateway.log | tail -1

Expected output: openclaw-mem0: initialized (mode: open-source, user: ..., autoRecall: true, autoCapture: true)


Modes

Mode Config Requires
platform apiKey from app.mem0.ai Internet, Mem0 API key
open-source oss block (self-hosted) Ollama + Qdrant (or other providers)

Configuration

Minimal Config

Platform:

"openclaw-mem0": {
  "enabled": true,
  "config": { "mode": "platform", "apiKey": "${MEM0_API_KEY}", "userId": "your-id" }
}

Open-source:

"openclaw-mem0": {
  "enabled": true,
  "config": {
    "mode": "open-source",
    "userId": "your-id",
    "oss": {
      "embedder":    { "provider": "ollama", "config": { "model": "bge-m3:latest", "baseURL": "http://localhost:11434" } },
      "vectorStore": { "provider": "qdrant", "config": { "host": "localhost", "port": 6333, "collection": "memories", "dimension": 1024 } },
      "llm":         { "provider": "ollama", "config": { "model": "llama3.2", "baseURL": "http://localhost:11434" } },
      "historyDbPath": "/absolute/path/to/.openclaw/memory/history.db"
    }
  }
}

Always set historyDbPath to an absolute path. When openclaw runs as a LaunchAgent, process.cwd() is /, so the default relative "memory.db" resolves to /memory.db (unwritable on macOS), causing a SQLITE_CANTOPEN crash loop. See troubleshooting.md.

Key Config Options

Key Default Notes
autoRecall true Inject memories before each agent turn
autoCapture true Store memories after each agent turn
topK 5 Max memories injected per turn
searchThreshold 0.5 Min similarity score (0–1)
userId "default" Scope memories per user

CLI

openclaw mem0 stats                          # Total memories, mode, user
openclaw mem0 search "user's name"           # Semantic search
openclaw mem0 search "topic" --scope long-term   # long-term | session | all

Agent Tools

The plugin registers 5 tools for agents to call:

Tool Description
memory_search Semantic search (scope: session/long-term/all)
memory_list List all memories for a user
memory_store Explicitly save a fact (longTerm: true by default)
memory_get Fetch memory by ID
memory_forget Delete by ID or query

Verifying It Works

# Check gateway log for injection events
grep "openclaw-mem0: inject" ~/.openclaw/logs/gateway.log | tail -5

# Confirm history DB created at configured path
ls -la /path/to/history.db

# No SQLITE errors since last gateway start
grep "SQLITE_CANTOPEN" ~/.openclaw/logs/gateway.err.log | tail -3

Troubleshooting

See references/troubleshooting.md for:

  • SQLITE_CANTOPEN crash loop (most common issue)
  • Qdrant / Ollama not reachable
  • Memories not being stored or recalled
  • Telegram allowFrom config warning