Skip to content

Installation

Requirements

  • Python ≥ 3.10
  • Git (Arbor runs each experiment in an isolated git worktree)
  • An API key for at least one LLM provider (Anthropic, OpenAI, or any OpenAI-compatible endpoint via LiteLLM)

Install

git clone https://github.com/RUC-NLPIR/Arbor.git
cd Arbor
pip install -e .          # or: uv pip install -e .

That single command installs Arbor and the arbor command into your current Python environment. We recommend a virtual environment so it stays isolated:

python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -e .

uv venv
source .venv/bin/activate
uv pip install -e .

Why editable (-e)?

Arbor is research software under active development. An editable install lets you pull updates with git pull without reinstalling.

Verify

arbor version
arbor doctor      # checks PATH, venv leakage, git, and API keys

arbor doctor is the fastest way to catch a broken setup — it reports which arbor your shell resolves, which Python it runs on, whether git is available, and whether your user config exists.

Optional: a global arbor command with pipx

If you'd rather have arbor available in every directory without activating a venv, install it with pipx — it manages the isolated environment for you:

pipx install -e .                 # run from the cloned Arbor directory
pipx reinstall arbor     # upgrade later

Troubleshooting

arbor: command not found

The package was installed into an environment that isn't active or on your PATH. Activate the right virtual environment, or use the pipx install above. Run arbor doctor for a diagnosis.

Next steps