OpenMetadata

OpenMetadata is an open-source data catalog and governance platform that has repositioned itself around AI agent enablement. The core product handles the usual data-catalog responsibilities — discovery, lineage, classification, quality monitoring, glossaries — across 130+ connectors covering Snowflake, BigQuery, Databricks, Airflow, dbt, Tableau and similar. What makes it interesting in 2026 is the Semantic Context Graph layer and a native Model Context Protocol (MCP) server, which together let LLM agents query trusted, governed metadata about an organization's tables, columns, metrics and business terms instead of guessing from raw schema names.

It's aimed at data platform teams at companies large enough to have a real catalog problem (millions of assets, dozens of sources) but that also want their internal AI agents and copilots to ground answers in something other than a vector dump of Confluence. Self-hosted is free and genuinely usable — 14k+ GitHub stars, 450+ contributors — while Collate, the commercial SaaS run by the same team, offers a managed cloud version with a free tier. There's also a published AI SDK for programmatic agent access.

The trade-off is operational weight: this is enterprise catalog software, not a lightweight RAG indexer. Expect non-trivial deployment, ingestion pipelines per source, and a learning curve around the metadata model. If you don't already need a catalog, bolting one on just to feed an agent is overkill.