Schema

Data Language Model: AI that understands business data directly. It takes tables and databases as they are, across any number of sources, and returns what the data is, how the sources relate, what is missing, and what can be predicted from it. No shared keys, no labels, no preprocessing step.

Built for agentic workflows that need data understanding. Agents meet new sources at runtime: a partner file drop, a schema change in production, a third-party API response, a database from an acquisition, a new customer onboarding. Schema understands that data in a single call, on the data as it stands, so the agent can act on it in the same workflow.

With Schema connected, Claude can:

• Run Schema on raw tables in one call, passed inline or referenced from connected data, and read back a structured result.
• Create an endpoint from your data with a held-out report, then call it, refresh it when the data changes, or upgrade it to a newer base.
• Connect databases, cloud sources, and REST sources, sync new snapshots, pin datasets, and generate synthetic data for a cold start.
• Get the exact cell count and expected run time of a call before it runs, and read your current usage.
• List and read endpoints, datasets, reports, and jobs, and cancel a job in flight.

Schema understands data.

Requirements: a Schema account on a subscription plan. You sign in with your existing Schema login or API key when you connect.