Skip to content

Installation

Requirements

  • Python 3.10+
  • uv package manager
  • AWS credentials — for LLM-based metrics via Bedrock, and for persistence via DynamoDB/S3

pyproject.toml + uv.lock are the single source of truth for dependencies.

git clone <repo-url>
cd uaef
uv sync            # install everything from the lockfile

uv sync installs everything needed to run every metric and use every capability — all metric backends (RAGAS, DeepEval), every adapter, analysis and reporting, the Bedrock LLM judge, and the UAEFClient remote SDK. Nothing else to install.

The only optional add-on is a Jupyter kernel for notebook use:

uv sync --extra notebooks   # or --extra all, which is core + the notebook kernel

The capability extras are no-ops

integrations, langgraph, analysis, judge, client, and server are retained as aliases so older install commands keep resolving. Every capability already ships in the required core, so these extras add nothing.

From a distributed package

If you received UAEF as a .whl or .tar.gz:

uv pip install ./packages/uaef-0.2.0-py3-none-any.whl

For development

uv sync --extra dev        # adds pytest, black, ruff, mypy, hypothesis
uv run pytest

See Development for the dependency rules, the notebook output filter, and code style.

Next

Run your first evaluation →