Skip to main content
Bring your existing agent to Chronicle to test how it performs in a simulated world.

Register your agent

Use a Chronicle integration or the Agents API to register your agent. This records its instructions, model, and tools so you can track changes between versions.

Set up execution

Provide your agent’s code, dependencies, model credentials, and a command that starts it. Registration alone does not make the agent runnable. During an evaluation, your agent receives a task and credentials for its test world. It completes the task and returns an answer; scorers grade the result. See execution setup for the details.

Run your first evaluation

Confirm your agent appears in Agents, then select it in an evaluation with a task suite. Inspect the scores and recorded actions to see where it succeeded or failed. When you change your agent, register a new version and compare it against the previous one using the same tasks.

Run an evaluation

Choose a task suite, run your agent, and review its results.