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

# Computer Use API

> Trigger, watch and approve computer-use jobs, manage templates, keys and webhooks — 90 v1 endpoints, generated from the agent platform's own OpenAPI.

The Computer Use API delegates a whole task — "reconcile these three invoices in NetSuite",
"file this return in the carrier portal" — to an agent that operates software through its screen.
You trigger a job, watch it or wait for it, approve the steps that need a human, and read the
outcome and its evidence when it finishes.

|          |                                                                                                                                      |
| -------- | ------------------------------------------------------------------------------------------------------------------------------------ |
| Base URL | `https://api.computer.stateset.app/api/v1`                                                                                           |
| Auth     | `X-API-Key: <tenant API key>` — issued at `POST /api/v1/keys`. The dashboard's own routes use a bearer session; the v1 API does not. |
| Spec     | `GET /api/v1/openapi.json` — the platform publishes its own v1 document                                                              |

<Note>
  This reference is generated from the platform's own OpenAPI, produced from the FastAPI app in the
  repository. Everything here is the versioned **v1** surface: the platform publishes its own
  changelog at `GET /api/v1/system/changelog`, its limits at `/system/limits`, and the webhook event
  catalogue at `/system/webhook-events`, so a client can discover what it is talking to rather than
  trust a doc. The dashboard's own internal routes under `/api/*` are not part of this contract and
  are not listed.
</Note>

<Warning>
  `api.computer.stateset.app` is the production host the platform's MCP client defaults to. At the
  time of writing it answers `503` — treat the playground on these pages as documentation of the
  contract until the deployment is up.
</Warning>

## Where to start

<CardGroup cols={2}>
  <Card title="Preview, then trigger" icon="play" href="/api-reference/computer-use/jobs-preview-create">
    `POST /jobs/preview` shows what a job would do without running it; `POST /trigger` runs it.
    Use preview first — a computer-use job acts on real software.
  </Card>

  <Card title="Wait for the result" icon="hourglass" href="/api-reference/computer-use/jobs-by-get">
    Poll a job, or subscribe to a webhook and let the platform tell you. The outcome carries its
    evidence — screenshots and the action log.
  </Card>

  <Card title="Approve what needs a human" icon="user-check" href="/api-reference/computer-use/approvals-list">
    Approval policies decide which steps pause for a person; the approvals endpoints are where
    that person says yes or no.
  </Card>

  <Card title="Make it repeatable" icon="layers" href="/api-reference/computer-use/templates-list">
    A template is a job definition you trigger by name, with its own limits and approval policy.
  </Card>
</CardGroup>

## The groups

| Group                                           | What lives there                                                                           |
| ----------------------------------------------- | ------------------------------------------------------------------------------------------ |
| **Jobs**, **Trigger**                           | Preview, trigger singly or in batch, read status and results, cancel, replay               |
| **Templates**, **Skills**, **Custom agents**    | Reusable job definitions, the skills a job may use, and agents with their own instructions |
| **Approvals**, **Approval policies**            | Which steps pause for a human, and the queue where a human answers                         |
| **Outcomes**, **Usage**, **Quota**, **Billing** | What was accomplished, what it consumed, and what remains                                  |
| **Webhooks**                                    | Subscriptions, the event catalogue, and a test delivery                                    |
| **Keys**, **Auth**                              | Tenant API keys                                                                            |
| **System**, **Health**                          | Discoverable metadata: version, limits, changelog, benchmarks                              |

## Related

* [Computer Use Agent](/computer-use-agent) — architecture, agent types, and how a job executes
* [Computer Use MCP server](/computer-use-mcp) — the same surface as nine agent tools, plus a local server
* [Computer-use engines](/computer-use-engines-mcp) — running under NSR governance