OpenPawz
Local AI agent that autonomously controls 25,000+ apps
Start new thread
Eli Bury

OpenPawz - Local AI agent that autonomously controls 25,000+ apps

by•
OpenPawz is a free, open-source AI desktop agent built with Rust + Tauri. Unlike cloud AI assistants, it runs locally — your data never leaves your device. What makes it different: it connects to 25,000+ integrations through an MCP bridge to n8n. Ask your agent to check Slack, create Trello cards, send emails, or query your CRM — it just does it. Key features: Any LLM provider (OpenAI, Anthropic, Google, Ollama) Zero-cost tool execution via local worker model MIT licensed, self-hosted

Add a comment

Replies

Best
Eli Bury
Maker
📌
Hey everyone! I built OpenPawz because I was inspired by what OpenClaw achieved — proving that a personal AI assistant running on your own machine is the future. But as I used it, I kept hitting friction: WebSockets everywhere — fragile connections that drop behind proxies, corporate firewalls, and Cloudflare. OpenPawz uses SSE streaming instead — one-way, reliable, works everywhere. Token costs spiral — Every tool call in OpenClaw burns tokens from your main model (Claude, GPT). Ask it to check Slack, send an email, and update Trello? That's three expensive API round-trips just for execution. OpenPawz solves this with the Foreman Protocol — a local 7B model on Ollama handles all tool execution for free. Your main model plans, the Foreman executes. Zero API cost for tool calls. Security concerns — credentials stored in config files and environment variables. OpenPawz stores everything in your OS keychain (macOS Keychain, Windows Credential Manager, Linux Secret Service). Keys never touch disk. Coding required for integrations — Want a new integration in OpenClaw? Write a skill in Python/TypeScript. OpenPawz connects to 25,000+ services through n8n's MCP bridge — no code, just point and click. Every n8n node is instantly available to your agent. Two innovations I'm proud of: The Librarian Method (Tool RAG) — Most agents load every tool into context upfront, wasting tokens and confusing the model. OpenPawz starts lean — the agent discovers tools on demand via semantic search. Like asking a librarian for the right book instead of carrying the whole library on your back. Desktop-native architecture — Built with Rust + Tauri, not Electron, not a browser tab. Native performance, tiny footprint, and a real desktop app with its own UI — not a terminal or a Telegram chat. I have massive respect for what @steipete built with OpenClaw. It proved the category. OpenPawz takes a different approach — desktop-first, zero-cost execution, no-code integrations, and security that doesn't keep you up at night. It's open-source (MIT), works with any LLM provider, and is actively being developed. What would you automate first? Happy to answer any questions!