---
title: Agents
description: Configure, authorize, provision, and execute Copilot agent runtimes.
---

# Agents

Agents are external headless CLIs. Copilot keeps the runtime, model provider, model, credentials, and authorization policy explicit.

<CardGroup cols={2}>
  <Card title="Agent contract" icon="file-lines" href="/agents/execution-contract">
    Understand the runtime/model tuple and effective configuration.
  </Card>
  <Card title="Runtime selection" icon="microchip" href="/agents/runtime-selection">
    Compare OpenCode, Codex, and Cursor execution modes.
  </Card>
  <Card title="OpenCode" icon="code" href="/agents/opencode">
    Follow the currently validated headless runtime path.
  </Card>
  <Card title="Codex and OpenAI" icon="robot" href="/agents/codex-openai">
    Configure the Codex runtime and understand OpenAI authentication boundaries.
  </Card>
  <Card title="Cursor (WIP)" icon="wand-magic-sparkles" href="/agents/cursor-wip">
    Review the implemented integration and the gates still pending.
  </Card>
  <Card title="Model selection" icon="brain" href="/agents/model-selection">
    Select and authorize the qualified model.
  </Card>
  <Card title="CLI commands" icon="terminal" href="/agents/cli-commands">
    Review the exact provider-specific command shapes.
  </Card>
</CardGroup>

## Recommended path

<Steps>
  <Step title="Define the tuple">Set the runtime, model provider, and model explicitly.</Step>
  <Step title="Authorize the model">Configure exact provider and model allowlists.</Step>
  <Step title="Provision and authenticate">Follow [Security & Operations](/security-operations).</Step>
  <Step title="Verify before writes">Run a bounded, read-only verification or smoke test.</Step>
</Steps>

## Related documentation

- [Agent input reference](/agents/input-reference)
- [Failure behavior](/agents/failure-policy)
- [Bugbot](/bugbot)
- [Development](/development)