quai-atlas

A read-only desktop application for looking at quai's work: experiments, runs, results, data versions, annotations, and the provenance that links them. It reads quai's records, its local MLflow tracking databases and the data repository's Git history, and changes none of them. It doesn't run quai, DVC, training or inference.

Atlas runs on Linux, Windows and macOS.

Install

quai-atlas is distributed on PyPI and requires Python 3.12 or newer. Install uv once, then create an environment of its own. The example uses Python 3.12; uv downloads it if needed.

uv venv --python 3.12 .venv
source .venv/bin/activate        # Windows (PowerShell): .venv\Scripts\Activate.ps1
uv pip install quai-atlas
quai-atlas

Git must be on the PATH for the data repository's checks. quai, DVC, MLflow and napari aren't needed to browse existing records.

First steps

1. Open Roots and mappings…. Add the experiment folder (the one holding quai_experiment.json, or a folder of experiments) with Add experiments…, and the data repository's root (the folder holding .git and .dvc) with Add data repository…. Save.
2. Choose an experiment on the left. Its overview lists the runs, with their analyses below them. The lineage graph above the table shows source → derived runs, recorded dataset IDs and separate analyses. Double-click a graph card or click a table row to open the run's summary.
3. The summary starts with its provenance verdict: what Atlas could check, and what it couldn't. Related items leads to the run's sources and to what was made from it.

A copy moved from another machine may need a mapping from the recorded path to its new folder, in Roots and mappings…. Tutorial 15 of the quai tutorials, "Browsing experiments and data repositories with quai-atlas", walks through all of it: verdicts, configuration, tracking, data versions, annotations, moved copies and network shares.

Images and annotations

Atlas can hand an image or label to napari, and start the annotation workbench for a project. Install them in the environment Atlas runs in:

uv pip install "napari[all]" quai-workbench-napari

quai-workbench-napari is distributed on PyPI too.

What Atlas keeps, and what it never does

Atlas keeps its settings and caches in quai/quai-atlas under the platform's user configuration folder, never inside a folder you register. It reads only inside the registered folders, and writes nothing there: no record, no Git command that changes the repository, no MLflow write. Opening a file or folder hands it to another application, whose own behaviour applies.

License

Apache License 2.0; see the LICENSE file.