Skip to content

Artigraph

Artigraph is a Python library for building data products as versioned, fingerprinted artifacts — not just as task runs or loose files. You declare the artifacts your workflow should produce, connect them with producers, and Artigraph records the fingerprints and snapshot metadata needed to reproduce, inspect, and hand off a coherent version later.

Use Artigraph alongside the runner you already have. It is not a scheduler or orchestrator: scripts, cron, Airflow, Dagster, or another tool can decide when a workflow runs while Artigraph owns the data contract — types, storage, fingerprints, and snapshot lineage.

  • Try the playground

    The in-browser playground runs Artigraph in an isolated Pyodide filesystem so you can edit a small graph without installing Python locally.

    Open the playground

  • Run the local demo

    Start with Pandas, Parquet, SQLite metadata, and a snapshot-aware readback flow.

    uv run --all-extras --with duckdb python docs/examples/snapshot_parquet_duckdb/demo.py

What problem does this solve?

After a pipeline has run a few times, "which files make up this dataset?" often becomes guesswork. A glob can pick up stale partitions; a rerun can overwrite an output you still needed; a downstream tool can silently mix versions.

Artigraph makes snapshot membership explicit. A version is something you can tag and reload, not a directory you hope is clean.

What Artigraph gives you

  • Artifacts describe the data product: type, format, storage, annotations, and statistics.
  • Producers describe how artifacts are built from other artifacts.
  • Fingerprints give artifacts, producers, partitions, and graph snapshots stable identities.
  • Snapshots let you tag and reload a coherent version of a graph later.
  • Storage metadata records which concrete partitions belong to a snapshot, so ecosystem tools can read the right files without guessing from path layout.
  1. Get started by building a small graph and running the tested local demos.
  2. Check status and fit for what works today and where Artigraph belongs in a stack.
  3. Read the graph lifecycle, then the core concepts: artifacts, producers, and snapshots/fingerprints.
  4. Run the snapshot-aware Pandas + Parquet example.
  5. Use the curated API reference for the public objects used in those examples.