Philosophy

Agent-to-agent collaboration opens uses of LLMs that a single agent cannot cover. It also accentuates the ways those models fail: hidden assumptions, loops that compound them, work that cannot be traced.

guilds.run is the framework Cigale AI built from years of running LLMs so that collaboration stays tractable.

System thinking used to be work for people who already wrote software. With LLMs it is available to anyone with a need or an interest. The level of abstraction changed. The need to see and steer the system did not.

Deeply inspectable

What the model sees is what you see.

Models make hidden assumptions. In a loop those assumptions stack. The operator cannot design a long-running multi-agent workflow if the guts stay hidden.

Every run keeps the prompt the model saw, every tool call, every side effect, every error, and every memory change. Each of those can be inspected and traced back.

Deeply customizable

Each connection, each tool available to an agent, and the model that agent uses is chosen for that job.

The same guild can mix open-weight and hosted models. A run uses the organisation's key for its agent's provider.

Iterate till perfection

A multi-agent system that survives long-running real use is hard to get right on the first pass. The product is built for iterative refinement: change a prompt, a tool, a routine, and watch the next run.

Share the love

Not everyone wants to spend time on system thinking, or has an interest in designing a well-oiled guild. Recipes exist so a working configuration can move from the person who built it to the person who needs it.

If builders share real-world recipes, the same expertise can serve day-to-day work beyond programming.