# Inspect Steward

> An agent that supervises evaluations, monitoring progress, diagnosing problems, and adjusting runtime state on your behalf.


## Getting Started

- [Getting Started](https://meridianlabs-ai.github.io/inspect_steward/index.html.md): Automated execution and supervision of long-running evaluations.

## Using Steward

- [Workflow](https://meridianlabs-ai.github.io/inspect_steward/workflow.html.md): A Steward run end to end: the workspace, launch and the tend loop, agent and human decisions, notifications, signoff, log directories and the log store, options, and advanced topics such as pausing and inflight changes.
- [Concurrency](https://meridianlabs-ai.github.io/inspect_steward/concurrency.html.md): Matching sample concurrency to inference concurrency: how sandboxes, samples, tasks and model connections are bounded, how the automatic ramp finds the right level, worker processes, and Docker configuration.
- [Error Handling](https://meridianlabs-ai.github.io/inspect_steward/errors.html.md): How Steward handles errors in unattended runs: automatic retry, what becomes a decision, codifying decisions as standing policy, resolving errors before signoff, and stuck samples.
- [Scanners](https://meridianlabs-ai.github.io/inspect_steward/scanners.html.md): How Steward scans transcripts for scoring integrity problems: the built-in scanner, choosing a scan model, adding your own, and what a finding asks of you.
- [Agent Runbook](https://meridianlabs-ai.github.io/inspect_steward/runbook.html.md): The operating instructions Steward ships to the coding agent supervising a run, reproduced verbatim.

## Integrations

- [Inspect Flow](https://meridianlabs-ai.github.io/inspect_steward/flow.html.md): Running an Inspect Flow spec under Steward: what Steward overrides in a worker, the two declarations that do not survive, and scanners.
- [Inspect Hawk](https://meridianlabs-ai.github.io/inspect_steward/hawk.html.md): Running a Hawk eval set config under Steward: how concurrency is set, what a local Hawk run cannot do, and what fan-out costs at startup.

## Reference

- [Reference](https://meridianlabs-ai.github.io/inspect_steward/reference/index.html.md): Command line reference for the steward CLI.