Advanced Usage

Real-world workflows for CI/CD, monorepos, team setups, and custom automation.

Advanced Usage

Autopilot is designed to work well in day-to-day coding, but it also fits more complex setups.

CI/CD Integration

Autopilot supports structured output on several commands, which makes it easier to wire into scripts and automation.

• autopilot status --json
• autopilot doctor
• autopilot insights --format json

Use these in CI to check repository health, validate setup, or collect metrics without parsing terminal styling.

GitHub Actions

Common CI usage patterns:

• Run autopilot doctor in a workflow to fail fast on broken setup.
• Run autopilot status --json as a health check for local runners.
• Use autopilot insights --export csv or autopilot insights --export json to produce report artifacts for maintainers.

Team Workflows

For team repos, combine the built-in safety defaults with a shared config file:

{
  "blockedBranches": ["main", "master", "production"],
  "requireChecks": true,
  "notificationsEnabled": true
}

Recommended team practices:

• Keep protected branches blocked.
• Enable pre-commit checks for lint and tests.
• Use the dashboard to review queue depth and recent commits.

Monorepos

Autopilot works best when you scope watching to the repo root and use ignore rules for generated packages or build outputs.

Tips:

• Ignore package build artifacts like dist/, build/, .next/, and coverage folders.
• Use .autopilotignore for generated workspace packages.
• Keep workspace-level changes in separate commits when possible so AI commit messages stay focused.

Git Hooks

Autopilot can be paired with your existing hooks:

• pre-commit for lint and test checks.
• pre-push for branch-specific guard rails.
• commit-msg if your team enforces message format rules.

If hooks take a while, increase your debounce window so Autopilot waits for them to finish cleanly.

Multi-Repo Setups

When you manage several repositories:

• Keep one .autopilotrc.json per repo.
• Use separate terminal sessions for each watcher.
• Avoid sharing a single log directory across repositories.

Offline Mode

Autopilot can still handle local commits when network access is unavailable.

• Git commits continue to work without network access.
• AI commit generation only runs when a provider is configured and reachable.
• Leaderboard sync and remote push actions will gracefully fail or queue depending on your configuration.

Custom AI Providers

Autopilot currently supports OpenRouter, Gemini, Grok, and none.

For OpenRouter specifically:

• Set ai.provider to openrouter.
• Use ai.model: "default" to auto-select a free model.
• Optionally set AUTOPILOT_OPENROUTER_MODELS to pin your own approved free-model list.

Accessibility

The CLI and docs are designed to stay usable in plain terminals and keyboard-driven workflows.

• Keyboard shortcuts are limited and predictable.
• JSON output is available for tooling and assistive wrappers.
• Dashboard controls are keyboard-only and can be used without a mouse.

Localization

The docs can be translated by adding localized content pages and keeping the navigation structure aligned.

If you want to maintain a translation set:

• Keep headings stable.
• Avoid hard-coding locale-specific terminology into command names.
• Prefer short sections that are easy to mirror across languages.

Performance

On large repositories, performance depends on:

• How much of the tree is watched.
• How many files are ignored.
• How often the watcher syncs status and leaderboard data.

Keep generated directories excluded and prefer a smaller watch scope when possible.