> ## Documentation Index
> Fetch the complete documentation index at: https://docs.surfacearea.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Use cases

> Two recipes built on worlds - gate every pull request on a graded run, and hand each of your own users a sealed simulation seeded from their own data.

Two recipes. Each gives the commands, the file to copy, and the settings that matter.

## Evals in CI/CD

**Every pull request gets checked against the world before it merges.**

The workflow opens a session per scenario on the platform, points the agent under test at each session's URL, grades the end state, and fails the check when the mean reward is below your bar. A second job runs the world's own test suite, so a world that stopped behaving like the vendor cannot pass a broken agent.

Use this when the agent changes often and you want a score on every branch.

[Evals in CI/CD →](/use-cases/evals-in-ci)

## Simulations for your users, from their real data

**Your customers run their agents against a copy of their own systems, not yours.**

Real traffic goes in, redacted on the way in if the rows carry personal fields (redaction is off unless you ask for it). You get a world per tenant or a per-user task on a shared world, then a session per run and a score per user. Support can still find one user's records in a hashed world by looking them up from the plaintext they know.

Use this when other people build or test agents on your product and need a safe copy of their data to act on.

[Simulations from real data →](/use-cases/sims-from-real-data)

<Info>
  Both recipes assume a world already exists. If one does not,
  [Get Started](/get-started) creates one from a shipped connector in a single
  command.
</Info>

## What they share

* **Every session copies the version's snapshot on open.** Two runs never see each other's writes.
* **A version is pinned.** The run records the exact world version it faced, so its score stays interpretable later.
* **The grader reads state, not transcript.** A run passes because the world holds what it should at the end, not because the agent said the right words.

## Where to go next

* [What a world is](/worlds) for the object both recipes are built on.
* [Run many sessions at once](/worlds/parallel) for the parallelism both recipes eventually need.
* [Glossary](/glossary) for the six words used throughout.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.