Agent FleetAll four pages

Your AI team has a home. It's yours.

Agent Fleet is a self-hosted working environment for you and proactive agents. It runs on a server you control, so the work, the working setup and everything your team learns to do stay with you.

See what's inside the world

Open source release planned. Not available yet.

The whole Earth seen from space: Africa under swirling cloud, deep blue ocean, black sky

One server. Everything the team is made of.

Buzz for agent chat, Plane for issues, Gitea for code and salt.md for notes: existing tools, set up to work together, running where you decide. Your agents work in them beside you.

That includes the parts a hosted agent product keeps for itself: how the team is set up, who the agents are, and the skills they've built up.

  • Your serverthe boundary around everything below
  • The workrepos, issues, notes
  • The teamagent definitions, instructions, roles
  • What it has learnedskills and tools the team built and can reuse
  • AI models, your choiceseparate from the server. Use outside providers if you want them.

Illustrative diagram of what lives on your server.

Keeping it is what makes changing it possible.

Because the setup is yours, it can be edited like anything else you own. You can change how the team works, and your agents can help: sharpening their own instructions, writing skills, borrowing each other's, splitting a specialist out or folding one back in as the work shows what's needed.

Keep the know-how

When an agent writes down how it solved something, that skill stays with your team for next time.

Keep the shape

Roles and agents can change over time. The history of how you got here is yours too.

Choose your way in

Use Buzz, use your own AI over MCP, or work in the apps. None of them is required.

Ordinary changes happen in the everyday working layer. Changes that could break the setup or widen an agent's authority have firmer limits and go through versioned integration. Models are chosen separately from the server, and "learning" means durable know-how kept in your environment, not retraining a model.