When AI agents start drifting, how do you debug them?

Hey Product Hunt! 👋

We’ve been running agent workflows for years, long enough to know exactly how they break. One agent makes a wrong assumption, another takes it as fact, the notes absorb it, and the error compounds. More instructions don’t fix that. Seeing what actually happened does. So we built guilds.run: agent teams you can understand, debug and control.

Get started in a few minutes

1. Run it on your machine: git clone <https://github.com/CigaleAI/guilds.git> && cd guilds && ./setup.sh && docker compose up --detach --wait

2. Open http://127.0.0.1:8080, connect your LLM provider, and set up a guild of agents, each with its own task and model.

3. Give them a real task and inspect what happens: the exact prompt, every tool call and result, every memory change, and the group chat between agents.

4. When something goes wrong, ask Studio why. It reads the runs, finds the cause and drafts a fix. Nothing changes until you publish.

Install guide: docs.guilds.run/getting-started/install

The goal isn’t just to make agents work. It’s to make them understandable enough that you can trust them with more.