Skip to main content
A twin is a running simulation of one service with its own records, API endpoint, and token. Your agent talks to it exactly as it would talk to the real service. Writes stick, and Chronicle records every request and every changed record.

​
Catalog

Find a service by name, resource, or the workflow you want to test. Expand any twin for its overview, vendor API, key resources, and an example evaluation with seed data and success criteria. Worldsmith twins can join a generated world with shared people and linked records. The Twins API lists the models and exact service IDs available to your deployment. Try the calls your agent needs against a running twin before building a large test suite.

​
Use a twin

In a launched world, open Twins and select a running service.
  1. Configure your API client with the twin’s base URL and API token.
  2. Read or update records using the service’s normal API paths and the IDs returned by the twin.
  3. Open Worldsmith → Timeline to inspect requests, responses, and state changes.
For a standalone twin, open Twins → Spin up twin, choose a service, and wait for Running. See the Twins API for the same steps in code.

​
Reset or stop

Reset throws away runtime changes and restores the original seed. Pause your agent first, reset, and wait for the twin to show Running again. Stop shuts it down. Twins also stop on their own when their lifetime (default 72 hours) ends.

​
Twins in evaluations

The twins you see in the dashboard are for exploring. When you run an evaluation, every trial gets its own fresh twins built from the saved world, and Chronicle hands your agent their URLs and tokens as environment variables. Nothing you did in the dashboard twins carries over, and nothing one trial does affects the next. See Connect your agent for how the agent receives those credentials.