Skip to content

Installation#

Two things, in order: the Python package, and a way to actually talk to a model. Both are one command each.

1. The package#

Requires Python 3.10+. f3dasm comes along automatically from PyPI.

pip install "adda @ git+https://github.com/bessagroup/adda.git"

(Not on PyPI yet, so it installs straight from the repository.)

2. A model to drive the agents#

By default adda talks to models through the Claude command-line tool, a separate program, not part of the pip install in the preceding step.

npm install -g @anthropic-ai/claude-code
claude   # first run prompts you to log in (subscription or API key)

That's it: once claude works on its own from your terminal, adda can use it. No Node.js? Or already have an API key and want to skip the login prompt?

export ANTHROPIC_API_KEY=sk-...

Don't have Claude access at all, or want to use a different model? adda also drives Ollama, any OpenAI-compatible endpoint, and vLLM: see Use a different model or backend.

Optional extras#

  • adda[viewer] adds the live run viewer (python -m adda.viewer <study-dir>), for watching a run as it works. See Watching a run.
  • adda[extra] adds docling for layout-aware PDF parsing in the literature reviewer (pulls torch; excluded on Intel macOS).
  • adda[docs] installs the documentation toolchain.
  • adda[tests] installs the test toolchain.
  • adda[dev] installs pre-commit and ruff.

Next#

Run the Quickstart: a couple of minutes to your first result.