Automation

    Move a ticket forward, and get a pull request back

    With Automation on, Planner can implement a ticket by itself. When a ticket moves into a state you have marked to trigger automation — typically In Progress — Planner runs a coding agent — Claude Code or OpenAI Codex, with your own key — against the project's GitHub, GitLab or Bitbucket repository and opens a pull request for your team to review. Nothing to install, no local checkout, no terminal.

    What Happens in a Run

    1. Planner checks the ticket can be implemented and records why if it can't
    2. It clones the project's repository into a temporary, isolated directory
    3. The agent works from the ticket's implementation brief: the description plus every comment, oldest first, so it builds what was agreed rather than what was first written
    4. If the agent produced a commit, Planner pushes a branch named after the ticket
    5. It opens a pull request (a merge request on GitLab) and links it in a comment on the ticket
    6. The ticket moves to your review state, if the project has one
    7. When your team merges the pull request, Planner notices within minutes, comments on the ticket and can move it to a state you choose, such as Done
    8. The temporary directory is deleted, whatever the outcome

    Built to Keep a Human in Charge

    • ✓ Never pushes to your main branch — every change is a pull request you approve
    • ✓ Never retries on its own, so one ticket never produces two pull requests
    • ✓ Every run is listed in the project's settings, including skipped runs and why
    • ✓ Runs have a time limit and a per-workspace concurrency cap
    • ✓ Tickets from a public form, email or chat always wait for a project admin to approve the run before the agent reads them
    • ✓ Optional plan-first mode: the agent posts a read-only plan on the ticket, and nothing is written until an admin approves it
    • ✓ Optionally, only tickets carrying a label you choose (e.g. "agent") run

    What the Agent Can and Can't Reach

    • ✓ It gets your AI model key and a copy of the repository — nothing else
    • ✓ It never sees your code host's access token: that is removed from its copy before it starts, and the push is made afterwards from a separate copy it never touched
    • ✓ None of our server's own credentials are passed to it
    • ✓ It is told that ticket text is a description of a change, not instructions to follow

    What Reviewers Get

    Every pull request Planner opens includes:

    • ✓ The agent's summary of what it changed and why
    • ✓ The tests or checks it ran, and their results
    • ✓ The approved plan, when the workspace plans first
    • ✓ How many files and lines changed
    • ✓ Warnings when the change touches CI workflows, dependencies, database migrations, container builds, environment files or security-sensitive code, or is unusually large

    Setting It Up

    Automation is included in the Business and Enterprise plans and is configured by a workspace Admin:

    1. Connect your code host: install the Planner GitHub App, or sign in with GitLab or Bitbucket — or paste an access token, which also works for self-managed GitLab
    2. Choose the agent — Claude Code or OpenAI Codex — and add your own key for it. Keys can be replaced or cleared, but never read back
    3. Pick a repository and base branch for each project
    4. Mark which workflow state triggers a run, and optionally a review state
    5. Choose your safeguards: plan-first mode and an optional required label
    6. Optionally, pick the state a ticket moves to once its pull request is merged
    7. Switch Automation on

    Your Code Host, Your Agent

    On GitHub, Planner uses a GitHub App instead of a long-lived personal token: it is installed on the repositories you choose, and each run uses an access token that expires within an hour. You can revoke it from your GitHub settings at any time.

    On GitLab and Bitbucket, the quickest way is to sign in and approve access; Planner renews that access by itself. For tighter control, use a project or workspace access token limited to the repositories Automation should reach instead. Either is stored encrypted and never shown again. Either way you pay your AI provider directly — there are no Planner AI credits.

    Prefer to run the agent yourself? Connect your coding tool through Studio and it can ask Planner for the same implementation brief.