Data Stack Index / v 02.06
Verified 2026·07·03
Send a correction
Catalog & discovery · primary Self-hosted only Open source

Amundsen.

LF AI & Data Foundation
Founded 2019
Status · ● active
Verified · ● 6d ago

The Lyft-born OSS catalog that invented search-first discovery — historically important, but development has largely stalled since 2024.

Capability profile Quality 0/3 Catalog 3/3 primary Lineage 1/3
Annual cost
Open core free to self-host · paid managed tiers
Deployment Self-hosted only
License Open source
Free tier Fully open source under Apache-2.0; no paid tier exists. Run cost is your own infrastructure (frontend, search, and metadata services plus Neo4j and Elasticsearch).
dbt integration Plugin
Persona data engineer · platform engineer
Company size scaleup → mid market
Warehouses snowflake · bigquery · redshift · databricks +5
OpenLineage consumer
01
Verdict

Where it fits — and where it doesn't.

● Ideal for

Teams that already run Amundsen and need to understand what they have, or teams with a genuinely minimal requirement — usage-ranked table search and ownership tracking, nothing more — who are comfortable owning a codebase that is no longer moving.

The core idea still holds up: index tables, dashboards, and people into Elasticsearch, rank results by query-log usage so the tables analysts actually trust float to the top, and keep the UX focused on the single question "which table should I use?" Databuilder's pull model is plain Python, so extending it from an existing Airflow deployment is straightforward for a data platform team.

○ Avoid if

You are choosing a catalog fresh in 2026 — the ecosystem has decisively moved to `datahub` and `openmetadata`, both of which cover Amundsen's discovery core plus glossary, governance, column-level lineage, and quality, with active communities.

Avoid also if you need any governance surface at all (glossary, PII classification, contracts, approval workflows — none exist), if you want a managed offering (Stemma is gone; no managed Amundsen exists), or if you can't staff the Neo4j + Elasticsearch + three-microservice operational footprint for software that won't be gaining features.

02
Strengths & weaknesses

The honest scorecard.

  • [+] Pioneered usage-ranked, search-first discovery — PageRank-style ranking from query logs remains a genuinely good idea that successors copied
  • [+] Deliberately small surface area — analysts get table search, owners, column stats, and previews without a governance platform's learning curve
  • [+] Apache-2.0 under neutral LF AI & Data governance, with nothing held back for a paid tier
  • [+] Databuilder is plain-Python ETL — custom extractors are easy to write and schedule from an existing Airflow deployment
  • [+] Battle-tested at real scale by Lyft, Square, Asana, Brex, and others during its 2019–2023 heyday
  • [−] Development has effectively stalled — last release August 2024 (databuilder 7.5.1), last monorepo commit April 2025 (moving a maintainer to emeritus)
  • [−] No managed offering exists — Stemma, the commercial managed Amundsen, was acquired by Teradata in 2023 and discontinued
  • [−] Discovery only — no business glossary, governance workflows, data contracts, or PII auto-classification
  • [−] Lineage is a table-level bolt-on (OpenLineage file extractor plus a lineage tab), far behind DataHub and OpenMetadata column-level lineage
  • [−] Heavy operational stack for frozen software — Neo4j, Elasticsearch, and three services to run and patch yourself
  • [−] Community momentum moved to DataHub and OpenMetadata; both publish well-trodden Amundsen migration guides
03
Editorial

What Amundsen actually is.

What Amundsen actually is

Amundsen is the data catalog Lyft built internally and open-sourced in October 2019, and the project that made search-first discovery the default shape of the category. The core insight — rank search results by usage from query logs, so the tables people actually query float above abandoned ones — was genuinely novel in 2019, and every catalog since has copied it. Architecturally it is a set of microservices: a Flask/React frontend, a search service backed by Elasticsearch, a metadata service backed by Neo4j, and Databuilder, a plain-Python ETL framework that pulls metadata from warehouses, dashboards, and dbt on a schedule (typically as Airflow DAGs). It joined LF AI & Data as an incubation project in August 2020.

The maintenance question

This is a know-what-you’re-getting-into entry. The honest state in mid-2026: the repository sits at roughly 4,800 stars and 216 historical contributors, but the last release was databuilder 7.5.1 in August 2024, commits through late 2024 were mostly dependency bumps plus one search-logging patch from Square, and the final commit to the monorepo — April 2025 — moved a maintainer to emeritus status. The repo is not archived and nothing stops you from running it, but no one should plan around new features, and security patches are best assumed to be your own responsibility.

The commercial story ended earlier. Stemma, founded by an Amundsen co-creator to sell a managed cloud version, was acquired by Teradata in 2023 and discontinued as a standalone product. Since then there has been no managed Amundsen, and both OpenMetadata and DataHub publish step-by-step migration guides aimed squarely at Amundsen installations.

Where it fits against the alternatives

Against datahub and openmetadata, the comparison is mostly historical: both cover Amundsen’s discovery core, then add business glossary, governance workflows, column-level lineage, and data quality — with active communities shipping monthly. Amundsen’s remaining edge is smallness: if your entire requirement is usage-ranked table search with owners and column stats, Amundsen does that with less conceptual overhead than either successor. Against apache-atlas, Amundsen is the friendlier UX but the less-maintained codebase; Atlas persists inside Hadoop-legacy estates Amundsen never targeted. Against select-star or secoda, the trade is self-hosted OSS versus paying for a maintained SaaS that does the same search-first job.

How to evaluate it

If you are evaluating catalogs fresh, evaluate DataHub and OpenMetadata first — either is the stronger default, and both are free to self-host. Amundsen is worth serious evaluation in exactly two situations: you have an existing deployment and are deciding when (not whether) to migrate, or you want the smallest possible discovery layer and are staffed to own a frozen Apache-2.0 codebase outright. In the first case, run the OpenMetadata or DataHub migration guide against a staging copy and time the effort. In the second, be honest about the operational bill: Neo4j, Elasticsearch, and three services patched by you, indefinitely, for software that will not improve.

04
Capability spec

All capabilities by cluster.

Catalog & discovery

Primary · strength 3/3
01 Business glossary — not supported
02 Glossary linked to assets — not supported
03 Natural language search — not supported
04 Ownership tracking
05 Data contracts — not supported
06 Governance workflows — not supported
07 Access request workflow — not supported
08 PII auto-classification — not supported
09 Tag propagation — not supported
10 Free self-hosted
Metadata ingestion pull connectors
Search approach keyword
Asset types tables · columns · dashboards · ml features

Lineage & metadata

Secondary · strength 1/3
01 Cross-system lineage — not supported
02 Upstream source lineage — not supported
03 Impact analysis — not supported
04 Reverse impact analysis — not supported
05 Historical lineage — not supported
06 Lineage API
07 Lineage diff — not supported
Granularity table level
OpenLineage consumer
Extraction openlineage events · api push
05
Warehouses & integrations

Where it plugs in.

Native warehouse support

snowflakebigqueryredshiftdatabrickspostgresmysqlmssqltrinoathena

Orchestrators & pipeline tools

airflowdbt-core
01dbt — Plugin
02Airflow — Plugin
03OpenLineage — consumer
04API access — full
05Terraform provider
06Public SDK — python
06
Pricing

The honest pricing breakdown.

Pricing model free forever
Charged per custom
Published ● Yes — listed on vendor site
Starts at $0 custom
Free tier ● Yes
OSS self-host ● Available

Free tier Fully open source under Apache-2.0; no paid tier exists. Run cost is your own infrastructure (frontend, search, and metadata services plus Neo4j and Elasticsearch).

Full Amundsen pricing breakdown — model, cost factors, alternatives by price →

07
Notable missing

What it doesn't do.

Business Glossary →

A managed vocabulary of business terms ("Active Customer", "Recognized Revenue") with definitions, owners, and — critically — links to the physical assets that implement them. Without the linking layer a glossary is just a wiki. With it, you can answer "which dashboards use our official definition of Active Customer?" — the question governance teams actually care about.

Column-Level Lineage →

Traces data flow at the individual column granularity rather than just between tables. Critical for impact analysis when a column changes, for PII tracking, and for regulatory compliance in financial or healthcare contexts. Column-level lineage is computationally expensive and not all tools that claim "lineage" actually provide it — many stop at table level.

PII Auto-Classification →

Automatically identifies columns likely to contain personally identifiable information — email addresses, phone numbers, national IDs — through regex, name heuristics, or ML. Required for meaningful compliance workflows at scale. Quality varies: naive implementations produce heavy false-positive rates. Worth asking vendors about their accuracy benchmarks.

Data Contracts →

Explicit, versioned agreements between data producers and consumers specifying schema, semantics, SLAs, and breaking-change policy. Enforced in CI for producers and at consumption time for consumers. Distinct from schema validation alone — a contract captures intent, not just structure. Implementations vary wildly; many tools claiming "data contracts" offer only schema checks.

08
Strong at

Drill into one capability.

Other key features

09
Alternatives & migrations

If not Amundsen, then what?

Common alternatives

DataHub → Best-in-class column-level SQL lineage parser (SQLGlot-based, benchmarked at 97–99% accuracy on standard corpora) ↔ Amundsen vs DataHub
OpenMetadata → Highest connector count in the OSS catalog space (120+) — particularly strong on dashboards, ML, and pipeline systems ↔ Amundsen vs OpenMetadata
Apache Atlas → Classification propagation via lineage plus Apache Ranger integration — tag-based access control and data masking enforced on the data itself, not just documented in the catalog ↔ Amundsen vs Apache Atlas
Select Star → Automated column-level lineage parsed from warehouse query logs is the differentiator — proven at very large scale (Block's case study covers lineage automation at exabyte scale) ↔ Amundsen vs Select Star

Teams typically migrate to

See all 9 Amundsen alternatives, scored and compared →
10
Common questions

Quick answers.

Is Amundsen open source?
Yes. Amundsen is open source under the Apache-2.0 license, and can be self-hosted at no license cost. A paid managed tier is also offered.
How much does Amundsen cost?
Amundsen publishes pricing, starting around $0 custom. A free tier is available: Fully open source under Apache-2.0; no paid tier exists. Run cost is your own infrastructure (frontend, search, and metadata services plus Neo4j and Elasticsearch).
How is Amundsen deployed?
Amundsen is self-hosted — you run it in your own infrastructure.
Does Amundsen work with dbt and my warehouse?
It integrates with dbt via plugin. Amundsen supports snowflake, bigquery, redshift, databricks, postgres, plus 4 more.

More catalog & discovery tools

Provenance.

Last verified 2026·07·03 against vendor documentation and, where possible, hands-on trial. Spot something off? Send a correction →

No paid placementNo vendor submissionsRankings never for sale Independence policy →