n8n-claw — Self-Hosted AI Agent

n8n-claw is a self-hosted, autonomous AI agent built on n8n, PostgreSQL, and Anthropic's Claude. It targets users seeking a customizable, integrated AI assistant for task management, knowledge retrieval, and automation, deployable via a single script for complete infrastructure control.

How It Works

The architecture uses n8n for workflow orchestration, connecting Telegram to an n8n-claw Agent (Claude Sonnet). It features modular components for task management, RAG-powered memory (with optional semantic search), expert agent delegation, and MCP skill integration. Background workflows automate memory consolidation and proactive reminders, emphasizing extensibility for custom API integrations and sub-agent delegation.

Quick Start & Requirements

• Installation: Clone repo and run ./setup.sh.
• Prerequisites: Linux VPS (Ubuntu 22.04/24.04 recommended) with min 4GB RAM, 15GB Disk. Requires Telegram Bot token/Chat ID, Anthropic API key. Domain name recommended for HTTPS.
• Dependencies: Docker, PostgreSQL, n8n, SearXNG, Crawl4AI. Optional: OpenAI API key (voice/photo), embedding API keys (OpenAI, Voyage AI, Ollama) for semantic search.
• Links: https://github.com/freddy-schuetz/n8n-claw

Highlighted Details

• Fully self-hosted AI agent using n8n for automation.
• Adaptive RAG memory with optional semantic search.
• Expert agents for specialized task delegation (Research, Content, Data Analysis).
• Extensible MCP Skills for custom API integrations.
• Proactive task management, reminders, and media understanding (voice, photos).

Maintenance & Community

Maintained by freddy-schuetz. Community support via GitHub Discussions. No specific sponsors listed.

Licensing & Compatibility

MIT License. Generally permissive for commercial use, subject to license terms. Requires external API keys.

Limitations & Caveats

External API keys are mandatory for core functionality (Anthropic) and advanced features (OpenAI for media, embedding providers for semantic search). MCP skill credentials are stored unencrypted in the database, posing a security risk if the VPS is compromised; encryption is planned. Some background workflows lack full abstraction for alternative LLM providers.