Amundsen

Overview

Amundsen is no longer maintained. Its repository was archived in September 2026 with the note that the contents remain available for historical purposes, and LF AI & Data now lists it under archived projects. The last release of any component was in August 2024, roughly two years before archival. Its domain has also lapsed and now redirects elsewhere, so the only safe link is the GitHub archive.

It is listed here because it still turns up in architecture diagrams and migration discussions, and because knowing it is dead is more useful than not finding it at all.

What it was: a data discovery tool built at Lyft and open-sourced in 2019, aimed at helping analysts, scientists and engineers find and trust tables, dashboards and streams. Its signature feature was search ranked by usage, in the style of page rank, so the tables people actually query surfaced first.

Architecturally it was a set of services: a Flask and React front end, an Elasticsearch search service, a metadata service on Neo4j or Apache Atlas, and databuilder, a Python library of extractors normally run from Airflow. It covered more than 26 sources, plus anything reachable through SQLAlchemy.

Its weak spot was governance. There was no business glossary, no automated classification and no policy engine, and column-level lineage was never a first-class feature, which is largely why the newer catalogues overtook it. If you are running it, OpenMetadata and others ship Amundsen connectors to help you move off.

Key features and capabilities

The same headings are used for every data catalogues & discovery entry, so two tools can be read side by side.

Metadata ingestion

• The databuilder Python library of extractors, run as batch jobs
• 26 or more documented sources, plus anything exposing a SQLAlchemy or DBAPI interface
• Pull only, normally orchestrated with Airflow

Search and discovery

• Elasticsearch search across tables, dashboards, streams and people
• Ranking by usage patterns, the feature it was best known for
• Column statistics and data previews on asset pages

Lineage

• Table-level lineage through the graph model
• Column-level lineage was never a first-class feature
• No lineage inference from query logs

Glossary, classification and policy

• Tags, badges and ownership
• No business glossary, automated classification or policy engine
• No data contracts or certification workflow

Ownership and collaboration

• People profiles showing owned and frequently used assets
• Editable descriptions on tables and columns
• Dashboards linked to the tables behind them

Integrations and APIs

• Warehouses and engines: Hive, Presto, Athena, Redshift, BigQuery, Snowflake, PostgreSQL, Delta Lake
• BI: Superset, Tableau, Mode, Redash, Databricks SQL
• dbt and Airflow, plus REST APIs on the metadata and search services

How it runs

• Self-hosted only; there was never a hosted service
• Docker Compose quickstart, community-contributed Kubernetes manifests
• Backends: Neo4j or Apache Atlas, with Elasticsearch for search
• Required Python 3.8 and Node 12, both now past end of life

Pricing

Open source

Free under the Apache 2.0 licence, but archived: there is no commercial edition, hosted service or paid support, and no further releases are expected.

Vendor pricing page →

Demos and videos

• Amundsen demo: tables, dashboards and people Product demo · YouTube
• Amundsen channel Channel or playlist · YouTube

About LF AI & Data Foundation (archived)

Amundsen was created at Lyft and open-sourced in October 2019, then donated to the LF AI & Data Foundation, where it ran as an incubation project. Its status there is now archived. The code is Apache 2.0 licensed and the repository is read-only. No company ever offered a commercial or hosted edition, and none will now. More than 35 adopters were listed in its README, among them Lyft, Square, ING, Snap and Instacart.

github.com