Release Autopilot is available to customers on select plansRelease Autopilot is only available to customers on certain plans. To learn more, read about our pricing or contact Sales.
- Your team merges changes faster than your release process can apply controls to them.
- Engineers act as manual reviewers for every release, which decreases the velocity of changes shipped.
- Creating, progressing, and cleaning up a flag for every change is too manual a process for the volume at which agents produce code.
Prerequisites
Before you use Release Autopilot, you need:- The Release Autopilot product offering.
- The Admin or Owner role, or a custom role that allows the Release Autopilot actions. To learn which actions, read the Release Autopilot setup prerequisites.
- A GitHub account and private repositories, if you use the LaunchDarkly GitHub app.
- Access to an AI model, either through your own model provider key or through LaunchDarkly inference.
How Release Autopilot works

- Auto-classification. Reviews a pull request when it’s marked ready for review and recommends a flag and a release plan.
- Auto-flag. After you approve, creates the flag and wraps the change in it, keeping the flag off for customers until the release starts.
- Auto-release. Starts each configured environment according to its trigger and progresses environments through the release based on the supported execution mode.
- Auto-cleanup. Opens a pull request that removes the flag from your code once it passes your lifecycle checks.
- Auto-archival. After cleanup, archives the flag automatically or submits an approval request to do so, depending on the project’s auto-archive settings.
Features and where to use them
The following table describes each Release Autopilot feature and where you can use it:| Feature | What it does | Where you use it |
|---|---|---|
| Auto-classification | Recommends a flag and a release plan for each pull request. | GitHub app, agent skills |
| Auto-flag | Creates the flag and wraps the change in it. | GitHub app, agent skills |
| Auto-release | Starts the release after deploy and progresses it through your environments in order. | LaunchDarkly, Releases API, MCP server, CLI, agent skills |
| Release tab and flag overview | Shows the release plan for a flag, lets you pause and unpause a release, and links to the pull request. | LaunchDarkly |
| Releases page | Shows every release in one view. | LaunchDarkly |
| Auto-cleanup | Opens a pull request that removes the flag from your code. | GitHub app, LaunchDarkly, agent skills |
| Auto-archival | Archives the flag after it passes your lifecycle checks. | LaunchDarkly |
| Agent skills | Runs the same steps from your editor, before you open a pull request. | Claude Code, Cursor, and other agents |
- In the pull request. The LaunchDarkly GitHub app comments on each pull request and responds to
@launchdarklycommands. To learn which commands you can use, read the Release Autopilot FAQ. - In LaunchDarkly. Review a flag’s release on its Release tab, and manage lifecycle settings and Release Autopilot settings from your project.
- In your editor or agent. Use agent skills, the LaunchDarkly MCP server, or the CLI to run the same steps without leaving your workflow.
Supported configurations
Release Autopilot currently supports the following configurations:- The LaunchDarkly GitHub app works only with private GitHub repositories.
- Each repository maps to one project, and each pull request maps to one or more flags.
- You approve each flag and release plan before Release Autopilot creates them.
- Release Autopilot bypasses approval workflows.
Next steps
Set up Release Autopilot
Install the GitHub app, map repositories to projects, and opt in to each capability.
Release Autopilot onboarding skill
Review the onboarding skill for Release Autopilot.
Release Autopilot agent skills
Review available agent skills.
Release Autopilot FAQ
Find pull request commands and answers to common questions.