OpenAI's self-service AI data agent, built on OpenMetadata

Most teams putting AI agents on their data run into the same problem: context. An agent that doesn't know which table to trust or what each column means returns answers that are confidently wrong by orders of magnitude.

Learn how OpenAI's Data Productivity team built Kepler, an internal AI data agent serving 3,500+ employees, on OpenMetadata as its open context layer. They turned days-long data hunts into self-service answers in under 90 seconds. Inside this case study:

• How OpenMetadata grounds every answer in governed metadata (schemas, lineage, query history) so the agent finds the right table by meaning rather than keyword match
• The six-layer context model and compounding memory that cut repeat queries from 22 minutes to under 90 seconds
• How Kepler enforces permissions at every layer, from retrieval to chain-of-thought, across 70,000 datasets and 580+ petabytes a day