N8N Claw
OpenClaw-inspired autonomous AI agent built entirely in n8n. Adaptive RAG-powered memory, Skills via MCP templates, Expert Agents with delegated sub-agents, proactive task management, media understanding - self-hosted with one setup script
Last verified:
What is N8N Claw?
N8N Claw is a fully self-hosted AI agent built on n8n, PostgreSQL, and Anthropic's Claude that serves as an OpenClaw alternative. It talks to users via Telegram, Slack, Teams, or a custom app, builds its own tools via MCP skills, manages tasks and memory, and runs entirely on your own infrastructure with no cloud lock-in. The agent understands natural language, remembers context across conversations, creates API integrations, and proactively keeps you on track with reminders and background checks.
Key features include hybrid search memory (semantic + full-text + entity matching with time decay), a knowledge graph tracking people and companies, expert agent delegation with 100+ specialized sub-agents, 65+ MCP skills for CRM/Project Management/Payments/etc., web search via SearXNG, voice/photo/PDF processing, Google Services OAuth integration, browser automation with interactive 2FA support, and project memory for persistent markdown documents. The setup takes about 10 minutes with a single setup.sh script handling Docker, databases, and credentials.
N8N Claw targets privacy-conscious users, GDPR-compliant organizations, self-hosting enthusiasts, automation professionals, n8n users, developers who want AI on their infrastructure, and anyone seeking an autonomous AI assistant without cloud subscription lock-in. It supports 8 LLM providers out of the box including Claude, GPT, Gemini, Mistral, DeepSeek, and Ollama.
N8N Claw pricing
Pricing model: Freemium
n8n-claw is 100% free and open source under the MIT license. There is no subscription, no paid tiers, and no cloud service fees. You only pay for your own VPS hosting costs and any LLM API keys you choose to use (Anthropic Claude, OpenAI GPT, Google Gemini, Mistral, DeepSeek, OpenRouter, or Ollama for local use with no API key). Optional embedding API for semantic search can use OpenAI, Voyage AI, or Ollama. Voice message transcription requires an OpenAI API key for Whisper. All infrastructure (Docker, PostgreSQL/Supabase, n8n, SearXNG, Crawl4AI) runs on your server at no additional software cost.
N8N Claw pros
- 100% self-hosted on your own infrastructure
- No cloud lock-in or subscription fees
- GDPR-ready with full data ownership
- Open source under MIT license
- Setup completes in approximately 10 minutes
- Single setup.sh script handles Docker, database, and credentials
- Telegram, Slack, Teams, and custom app support
- Hybrid memory search with semantic + full-text + entity matching
- Knowledge graph automatically tracks people, companies, and relationships
- 100+ expert sub-agents across 12 categories
- 65+ MCP skills for CRM, project management, payments, banking
- Built-in web search via SearXNG with no API key needed
- Voice messages transcribed via OpenAI Whisper
- Photo analysis via OpenAI Vision
- PDF text extraction supported
- Google Services OAuth2 with automated auth flow via Telegram
- Browser automation with interactive 2FA support
- 8 LLM providers out of the box including Ollama for local use
- Project memory with persistent markdown documents
- Proactive heartbeat reminders for overdue tasks
- Smart background checks that only notify on new findings
- OpenClaw integration for autonomous coding tasks
- MCP Builder creates custom skills from natural language
- Recurring actions scheduled in natural language
N8N Claw cons
- Requires a Linux VPS with minimum 4GB RAM and 15GB disk
- Domain name needed for Telegram HTTPS webhooks (or use sslip.io)
- Telegram bot accepts messages from anyone unless manually restricted
- Skill credentials stored in plain text in PostgreSQL
- Voice messages require OpenAI API key for Whisper
- Semantic memory search requires separate embedding API key
- MCP-triggered executions limited to 5-minute timeout
- MCP interface supports text only, no binary inputs
- Supabase Studio only accessible via SSH tunnel
- Single-user optimized, may bottleneck with thousands of memory entries
Frequently asked questions about N8N Claw
What is n8n-claw?
n8n-claw is a fully self-hosted AI agent built on n8n, PostgreSQL, and Anthropic's Claude. It serves as an OpenClaw alternative that talks to you via Telegram, Slack, Teams, or a custom app. The agent manages tasks, remembers context across conversations, builds API integrations via MCP skills, sends proactive reminders, and runs entirely on your own infrastructure with no cloud lock-in. You control how agents operate, how data is managed, and how everything connects.
How do I install n8n-claw?
Installation requires one command on your VPS: git clone https://github.com/freddy-schuetz/n8n-claw.git && cd n8n-claw && ./setup.sh. The interactive setup script handles Docker, databases, credentials, and all services automatically. You'll need a Telegram Bot Token (from @BotFather), your Telegram Chat ID (from @userinfobot), an LLM API key, and a domain name (or use sslip.io). Setup takes about 10 minutes and guides you through choosing your LLM provider, agent personality, and optional features like embeddings and voice messages.
What LLM providers does n8n-claw support?
n8n-claw supports 8 LLM providers out of the box: Anthropic Claude (default), OpenAI GPT, Google Gemini, Mistral, DeepSeek, Ollama (local, no API key needed), OpenRouter, and any OpenAI-compatible endpoint. You choose your provider during setup with no manual configuration needed. Ollama is particularly useful for local deployment with no API costs.
How does the memory system work?
The memory system uses hybrid search combining semantic search (vector embeddings via pgvector), full-text search with accent/umlaut normalization, and entity match boosted by people/companies/projects. Results are fused via Reciprocal Rank Fusion (RRF) with time decay favoring recent memories. Memories are automatically tagged with keywords and linked to named entities. Category-based auto-expiry removes general memories after 90-180 days while contacts, preferences, and decisions never expire. Nightly consolidation at 3am summarizes conversations into long-term memory and extracts behavior patterns.
What are Expert Agents?
Expert Agents are specialized sub-agents that handle complex tasks. Each has its own persona, tool access, and independent Claude instance. Three experts come included by default: Research Expert (web research, fact-checking, structured summaries), Content Creator (copywriting, social media posts, blog articles), and Data Analyst (data analysis, pattern recognition, KPI interpretation). Over 100 more experts are available from the agent catalog across 12 categories: Analytics, Marketing, Development, Sales, Creative, Operations, Product, Research, Communications, Education, HR, and Leisure. The main agent automatically delegates to the right expert based on your request.
What MCP skills are available?
There are 65+ ready-to-install MCP skills in the catalog. Categories include CRM (HubSpot, Salesforce, Zoho), Project Management (Jira, Confluence, Asana), Payments & Invoicing (Stripe, Lexware), Banking (KontoFlux with 5000+ German banks), Spreadsheets (Airtable, NocoDB), Productivity (Notion, Todoist, GitHub, Google Calendar, Google Drive, Nextcloud), Email (Gmail API, IMAP/SMTP), Calendars (Google Calendar, CalDAV), Meetings (Vexa bot for Google Meet/Teams/Zoom), Marketing Analytics (Google Ads, GA4), Home Assistant, and 50+ more including Wikipedia, Hacker News, DeepL, YouTube, recipes, and TikTok Ads. Install with a single chat command like 'install the HubSpot skill'.
How do I integrate Slack or Teams?
For Slack: Create a Slack App at api.slack.com/apps, add bot token scopes (chat:write, channels:history, channels:read), install to workspace, create a Slack API credential in n8n, enable the Slack Trigger + Slack Reply nodes in the Webhook Adapter workflow, activate the workflow, add the webhook URL to Event Subscriptions, subscribe to message.channels and message.im events, and invite the bot to your channel. For Teams: Register a Bot in Azure Portal, create App ID + Client Secret, set messaging endpoint to the Teams Trigger webhook URL, create Microsoft Teams OAuth2 credential in n8n, enable Teams Trigger + Teams Reply nodes, and activate the Webhook Adapter workflow.
Is my data secure with n8n-claw?
Yes - n8n-claw is 100% self-hosted on your infrastructure. No data leaves your network except LLM API calls. It's GDPR-ready with full data ownership under MIT license. Your Telegram bot can be restricted to your chat ID only by adding Allowed Chat IDs in the Telegram Trigger node. PostgreSQL and PostgREST run on Docker-internal networks only, not publicly exposed. Supabase Studio is bound to localhost and accessible only via SSH tunnel. The main security concern is that skill credentials are stored in plain text in PostgreSQL, so secure SSH access with key-based authentication is essential.
Can I use n8n-claw without Telegram?
Yes - n8n-claw exposes an HTTP Webhook API at https://YOUR-DOMAIN/webhook/agent that allows any system to interact with the agent without Telegram. Send POST requests with message, user_id, and optional session_id/source/metadata fields. The Webhook Adapter workflow supports Slack, Teams, Discord, Paperclip, and custom integrations with input/output mapping. You can also connect via Claude Desktop, ChatGPT, Claude Code, Lovable, or Cursor using n8n's Instance-Level MCP Server with OAuth authentication. Discord requires the optional discord-bridge sidecar container.
What happens if workflows fail?
n8n-claw has built-in failure awareness - workflow errors are automatically logged to memory and pushed to Telegram as alerts. You can ask 'did anything fail today?' or 'what was the error in the last background check?' and the agent explains the error in plain language without needing to dig through n8n execution logs. The Error Notification workflow catches failures, sends Telegram alerts, and logs to memory_long for later review. This proactive error awareness keeps you informed without manual log monitoring.