AGENT SKILL

muninn-memory

暂无简介

License 许可
MIT
最近更新
2026-06-07
Stars
★ 0
Forks
1

03 SKILL.md 正文

Muninn Memory

A local-first memory system for AI agents. Store episodic, semantic, and procedural memories with intelligent retrieval.

Installation

clawhub install muninn-memory

Or install from source:

cd ~/.openclaw/workspace/skills
git clone https://github.com/openclaw/muninn-memory.git
cd muninn-memory
npm install && npm run build

Requirements

  • Node.js 18+
  • Ollama running locally (for embeddings)
  • SQLite (included)

Quick Start

import { MemoryStore } from 'muninn-memory';

// Initialize (creates SQLite DB)
const memory = new MemoryStore('./memories.db');

// Store a memory
const mem = await memory.remember(
  'Phillip lives in Brisbane, Australia',
  'semantic',
  { entities: ['Phillip', 'Brisbane', 'Australia'], salience: 0.8 }
);
console.log('Stored:', mem.id);

// Recall memories
const results = await memory.recall('Where does Phillip live?');
console.log(results[0].content);
// "Phillip lives in Brisbane, Australia"

// Get stats
console.log(memory.getStats());
// { total: 1, byType: { episodic: 0, semantic: 1, procedural: 0 }, entities: 3, edges: 0, procedures: 0 }

memory.close();

Memory Types

Type Use Case
episodic Events, conversations, experiences
semantic Facts, knowledge, relationships
procedural Skills, workflows, step-by-step processes

API

remember(content, type, options?)

Store a new memory.

await memory.remember(
  'Meeting with Sarah about Q3 roadmap',
  'episodic',
  { 
    entities: ['Sarah', 'Q3 roadmap'],
    salience: 0.7,
    title: 'Q3 Planning Meeting'
  }
);

Options:

  • title?: string - Short title
  • summary?: string - Brief summary
  • entities?: string[] - Extracted entities
  • topics?: string[] - Topics/tags
  • salience?: number - Importance 0-1 (default: 0.5)

Returns: Memory object with id, content, type, entities, etc.

recall(query, options?)

Search memories using hybrid search.

const results = await memory.recall('Sarah Q3', {
  limit: 5,
  types: ['episodic', 'semantic'],
  entities: ['Sarah'],
  topics: ['planning']
});

Options:

  • types?: MemoryType[] - Filter by type
  • entities?: string[] - Filter by entity
  • topics?: string[] - Filter by topic
  • limit?: number - Max results (default: 10)

Returns: Array of Memory objects sorted by relevance.

forget(id, hard?)

Delete a memory.

// Soft delete (can be recovered)
memory.forget('m_abc123');

// Hard delete (permanent)
memory.forget('m_abc123', true);

getEntities()

List all extracted entities.

const entities = memory.getEntities();
// [{ name: 'Phillip', memory_count: 5, last_seen: '2026-02-24...' }, ...]

getStats()

Get vault statistics.

const stats = memory.getStats();
// { total: 42, byType: {...}, entities: 15, edges: 8, procedures: 3 }

CLI Usage

# Run tests
node dist/index.js test

# Start MCP server (for agent integration)
npm run mcp

MCP Server

Start the MCP server for agent integration:

npm run mcp

Security Verification:

Claim Evidence
stdio-only transport src/mcp/server.ts:5 imports StdioServerTransport from @modelcontextprotocol/sdk/server/stdio.js — no HTTP/TCP listeners
Local-only network Only fetch('http://localhost:11434/...') for Ollama embeddings — no external servers
No postinstall scripts package.json has no postinstall, preinstall, or postbuild scripts
Dependencies Only better-sqlite3 (local DB), ollama (local embeddings), uuid — all from npm
Credentials No environment variables, no secrets in code

Database: Default location is ./openclaw-memory.db (SQLite, local file).

Known Limitations:

  • Memory content is passed to LLM prompts — users could inject instructions via stored memories
  • Calls local Ollama at localhost:11434 for embeddings — requires Ollama running

Security Notes

  1. ✅ No external network — MCP server uses stdio, not HTTP/TCP. Only calls localhost:11434 for local Ollama embeddings.
  2. ✅ No authentication required — Only accessible to the parent process
  3. ✅ Local-only database — SQLite file, no remote connections
  4. ✅ No credentials needed — Only requires local Ollama for embeddings
  5. ✅ No postinstall scripts — Clean npm install
  6. ⚠️ Prompt injection risk — Memory content is passed to LLM prompts. If users store malicious content, it could affect LLM behavior. This is inherent to all memory systems.

Database Location

Default: ./openclaw-memory.db

Custom path:

const memory = new MemoryStore('/path/to/custom.db');

Features

  • Hybrid Search — BM25 + semantic (embedding) with reciprocal rank fusion
  • Entity Extraction — Auto-extract people, orgs, projects, technologies
  • Spelling Variants — UK↔US English expansion (colour/color, etc.)
  • Question-Type Aware — Factual questions prioritize semantic memories
  • Entity Boost — Memories with matching entities rank higher

Benchmark

System LOCOMO Score
Muninn 93%
Engram 79.6%
Mem0 66.9%

Muninn Phase 1.5 includes knowledge graph integration for temporal reasoning and contradiction detection.

License

MIT — Free, open source, no restrictions.


🦜 Built by KakāpōHiko (KH)

04 同仓库更多 Skills

javimosch
javimosch/open-claw-skills

crowd-prompting

A marketplace where AI agents improve prompts, system instructions, tool descriptions, and other text-based content with domain expertise from real-world operations — and earn tokens for valuable contributions.

★ 0
javimosch
javimosch/open-claw-skills

ceaser

Interact with the Ceaser privacy protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- shield, unshield, note management, and protocol queries. All ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). Ceaser is a privacy-preserving ETH wrapper using Noir/UltraHonk zero-knowledge proofs on Base L2 (chain ID 8453).

★ 0
javimosch
javimosch/open-claw-skills

ceaser-send

Fully automated private ETH transfer via Ceaser Protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- all ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). WARNING -- creates observable on-chain links between user wallet and hot wallet (see Privacy Warning). Generates an ephemeral hot wallet with BIP-39 mnemonic recovery, user funds it, agent signs and broadcasts the Shield TX automatically. Exactly one manual step required (funding the hot wallet with ETH). Uses Noir/UltraHonk zero-knowledge proofs.

★ 0
javimosch
javimosch/open-claw-skills

dilemm-ai

Compete in DilemmAI, the prisoner's dilemma AI arena at dilemm.ai. Use when an OpenClaw agent wants to create an account, design and submit strategy prompts for their bot, enter matchmaking, analyze historical matches and leaderboards, and iterate on bot strategy. Requires headless browser automation (no API available yet).

★ 0
javimosch
javimosch/open-claw-skills

moldium

Post and manage content on the Moldium blog platform. Triggered by "post to Moldium", "write a blog post", "publish an article", etc.

★ 0
javimosch
javimosch/open-claw-skills

arxiv-paper-reviews

Interact with arXiv Crawler API to fetch papers, read reviews, submit comments, search papers, and import papers. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, submitting paper reviews, searching papers by title, or importing papers from arXiv URLs via API at http://weakaccept.top:8000/.

★ 0