Run tests & review results | Sarvam API Docs

A run plays every test (or a single test) against your agent, then grades each conversation.

Start a run

Run all tests or one test

Click Run all to run every test, or Run on a single test while you are writing or fixing it. Running all tests consumes more credits: every test executes, multiplied by the runs per test case.

The Run tests dialog with runs-per-test-case options and an optional run name. — Starting a run: choose how many times each test case executes, and optionally name the run.

Check the agent version

Tests run against the agent version open in the Canvas — switch versions from the selector in the top bar. See Agent versioning.

Choose runs per test case

Each test can execute 1, 2, 3, or 5 times. Conversations vary between executions, so running a test more than once shows whether a failure is consistent or occasional. Use 1× while authoring; raise it before a release.

Watch progress

A run simulates whole conversations, so it takes minutes. The progress bar shows completed executions, and each test shows its pass or fail status as it finishes.

Read the results

Open a run to see each test’s result — a test passes only when every expected behavior passed. Select a test to see:

The transcript — the full conversation between the simulated user and your agent.
Per-behavior verdicts — each expected behavior is graded Pass or Fail with a short explanation.

When a test ran more than once, each execution appears as its own tab (Run 1, Run 2, …).

Run results showing the list of tests with pass and fail markers, the conversation transcript, and each graded behavior with its explanation. — A run's results: per-test pass and fail, the transcript, and the verdict for each expected behavior.

Write test cases

Add or adjust the cases behind a failure.

Tests API

Create, run, and read suites over REST.