# While

> The whileai Python SDK: post-training data, evals and training loops for language-model agents.

- [whileai](https://docs.withwhile.com/index.md): Training and eval data for agents that call tools. Install, run the sixty-second offline example, then pick a path.
- [Evals for the agent you already have](https://docs.withwhile.com/evals.md): A pass rate with an interval, a table of where the agent fails, and a CI check that turns red when it gets worse. Offline, no key, seconds.
- [While Simulations](https://docs.withwhile.com/simulations.md): How the simulation engine thinks and why: the problem it solves, the situations it covers, and the rows it returns.
- [The engine on one page](https://docs.withwhile.com/engine.md): How simulate() makes evals and training data in eight steps: coverage, a sandbox world with failure modes, and a judge validated before training.
- [Character training with whileai](https://docs.withwhile.com/character-training.md): Change the weights so a model has a stable way of talking without a system prompt: sources, the recipe, and what to measure.
- [Reward hacking detection with whileai](https://docs.withwhile.com/reward-hacking.md): How the SDK looks for over-optimization before a run, during it, and after: the gap between training reward and the eval you care about.
- [Safety evals for tool-using agents with whileai](https://docs.withwhile.com/safety-evals.md): Safety evals for tool-using agents: private data, actions on state, and outbound sends, and whether they can be turned against their owner.
- [whileai reference](https://docs.withwhile.com/reference.md): Every call, knob, report and gate in one page, in the order a post-training run happens.
- [API reference](https://docs.withwhile.com/api/index.md): Every public name in whileai and whileai.simulations, with its signature and docstring, generated from the package.
- [whileai](https://docs.withwhile.com/api/whileai.md): The platform client: login, trace ingest, account.
- [whileai.simulations.constants](https://docs.withwhile.com/api/constants.md): Module-level constants and their defaults.
- [whileai.simulations.data](https://docs.withwhile.com/api/data.md): SimulationData, conversation rebuild, row export, and the grade
- [whileai.simulations.environment](https://docs.withwhile.com/api/environment.md): Export a simulation as an RL environment a trainer can install and drive.
- [whileai.simulations.export](https://docs.withwhile.com/api/export.md): Training-ready rows: system prompt, tool schemas, standard wire format.
- [whileai.simulations.generate](https://docs.withwhile.com/api/generate.md): Public names in whileai.simulations.generate.
- [whileai.simulations.ingest](https://docs.withwhile.com/api/ingest.md): Public names in whileai.simulations.ingest.
- [whileai.simulations.monitor](https://docs.withwhile.com/api/monitor.md): Is the run hacking its reward right now? Watch it while it trains.
- [whileai.simulations.run](https://docs.withwhile.com/api/run.md): The run engine behind simulate(): knobs, spec loading, the scheduler.
- [whileai.simulations.schema](https://docs.withwhile.com/api/schema.md): The typed row: four objects, one flat wire shape, one version stamp.
- [whileai.simulations.score](https://docs.withwhile.com/api/score.md): Public names in whileai.simulations.score.
- [whileai.simulations.simulation](https://docs.withwhile.com/api/simulation.md): `simulate()`: the public entry point.
- [whileai.simulations.training](https://docs.withwhile.com/api/training.md): Training runs: the loss curve and the progress bar on the platform.
- [whileai.simulations.verify](https://docs.withwhile.com/api/verify.md): Verifiers: programmatic, verifiable rewards (RLHF book ch. 7, 13).
- [whileai.simulations.world](https://docs.withwhile.com/api/world.md): Public names in whileai.simulations.world.