An octopus AI agent that wrangles other agents into reliable teams.
The open-source framework for AI agent teams — assemble agents that plan, execute, review their own work, and ping you only when it matters.
Why Polpo?
AI agents are great at execution. They’re terrible at finishing complex tasks autonomously. Without oversight they drift, conflict, and stall — and nobody tells you until it’s too late. You end up babysitting the machines that were supposed to save you time — 4 monitors, 12 terminals open, and zero confidence that anything actually works.Polpo fixes this. Eight arms, infinite patience, and a Telegram account. You talk to Polpo like you’d talk to a project manager — describe what you need, and it assembles the right agents, plans the work, checks every result, retries what’s broken, and escalates what it can’t fix. You define the goal, Polpo makes sure it actually gets done.
Who is it for? Anyone who wants a personal AI team on their laptop, a $5 VPS, or a Docker container in their cloud. Tell Polpo to build an app, run a virtual AI company, or automate your workflows — and it handles the rest. One npx polpo-ai and you’re running. No hosted platform, no vendor lock-in — your hardware, your API keys, your rules.
Prerequisites. Node 18+ and an API key from any of the 22+ supported providers. Run npx polpo-ai and the setup wizard will guide you through model and key configuration.
What’s different
- Lives its own life. Polpo builds missions, picks agents, and works through queued tasks on its own — 24/7 if you need it. Describe what you need and walk away.
- Reliable output, not best-effort. Polpo scores every task with LLM judges. Below threshold? It sends the agent back to fix it. You get the result, not the retries.
- Playbooks. Define a mission once, run it forever. Schedule it, tweak it, improve it. Your agent teams get better over time.
- Proactive, not a chatbot. You decide how and when Polpo should reach you — Slack, email, Telegram, webhooks — and it only does when it matters.
How it works
