adda

adda runs a team of LLM agents on a data-driven engineering problem. You describe the problem in one file, and adda hands back a notebook that reproduces the result end to end.

What do you want to do?

Get started

• Install adda and connect a model.
• Run the quickstart: a real problem, a real answer, a couple of minutes.
• Tour the viewer on a finished run.

Do a task

• Write a study.
• Watch a run and steer it.
• Understand what a run produced.
• Use a different model or backend.
• Change an agent's tools or build your own agents.
• Fix a run that failed or stalled.

Understand how it works

• The permission graph: who may delegate to whom.
• How a run is kept honest: the hypothesis ledger, the critic, and the reproduction gate.
• Features: how a piece of the scaffolding is turned off for an experiment.

Look something up

• Configuration reference and Runtime reference: every config.yaml key.
• Command-line reference.
• Python API reference.

adda builds on f3dasm for the data-driven primitives and adds the agentic orchestration on top.