Hey Product Hunt! 👋 I’m Robin, the creator of Myrmidon.

Myrmidon is a free, open-source AI agent platform for your PC. One agent, a whole crew, and every step accountable.

I built Myrmidon because I wanted an AI agent that could do more than answer questions or write code. I wanted one that could actually get things done: investigate problems, use a real browser, work with apps on my Mac, and bring in a team of specialist agents when a task gets complicated.

The idea is simple: you give Myrmidon a goal, and it figures out how to get the work done.

With Myrmidon, you can:

Delegate complex tasks to AI agents that plan, collaborate, and review each other’s work.
Code in your project with checkpoints, isolated Git worktrees, and reviewable diffs.
Automate a real browser and use approved apps on your Mac.
Chat with your agent from messaging apps and continue the same work across channels.
Choose from supported AI models, including signing in with subscriptions you already have.
Keep control through permissions, approval gates, an audit trail, and the ability to rewind changes.

For example, imagine discovering a bug in your checkout flow. Ask Myrmidon to investigate it, reproduce the issue in a browser, delegate the fix to a coding agent, and run the tests. You can follow the work, inspect the changes, and review the result instead of blindly trusting the final answer.

Myrmidon is open source under Apache 2.0, and I’m building it in public. It’s still in beta, and there’s plenty left to improve. I’d love to build this with people who care about capable AI agents that remain transparent and under your control.

The app currently targets macOS on Apple silicon. You’ll need access to a supported AI provider or API key, and provider usage limits and fees may apply.

🚀 Try Myrmidon: https://github.com/Robinbinu/myrmidon
💻 Source code: https://github.com/Robinbinu/myrmidon
🎬 Watch the demo: https://youtu.be/FnOd7sCxWWM

I’d love your feedback, especially on setup, the first task you try, and where the agent or multi-agent workflow falls short.

What would you delegate to a crew of AI agents if you could watch and verify their work?