- Preview means the product works today but may change.
- Requires enablement means someone must turn it on before it appears for your organization.
What is on, and what needs turning on
To see which apps your organization has, see Find your installed apps.
Find an enabled app
If Evals is missing from the app rail, open ⋯ (More). Select Evals, or pin it to the rail. If there is no ⋯ tile or Evals is not listed, ask your administrator about enablement. The guided Build an eval flow has separate enablement. Its absence does not prevent SDK evals or experiment review.What AI Registry does not do
- A record does not start an agent. Registering an agent or recording a session stores what happened. Your framework, runtime, or local coding client runs the agent. See How agents work.
- An accepted upload is not proof. A successful trace or session upload does not prove the run completed, or that it landed in the workspace you intended. Confirm it in the app or with a read. See Send and verify traces.
- A workspace ID is not access. What you can see comes from the identity you sign in with and its permissions, not from the ID a caller sends.
- Usage is not a quality verdict. A high run count shows reuse, not correct results. Use Evals to measure quality.
- MCP Gateway does not store tool data. For calls in a verified session, it can record the tool name, outcome, and duration for audit, but not the arguments or output. Recording is best-effort, and a failed audit write does not stop the tool call. See Track usage and audit tool calls.
- Metrics Studio and Ontology never run SQL. They store agreed definitions and context that agents read. Your agent or warehouse runs the query.