Your first agent pull request, in about ten minutes

1. Pick a runner and sign in

A runner is the program on your Mac that does the work when an issue is labelled. There are three; pick one.

The Celiador editor

The loop built into a VS Code fork, with an Agents view that shows every run. You need an Apple Silicon Mac (macOS 13 or later) or Linux x86_64. Download the macOS DMG or pick a Linux package on the download page. Open it, drag Celiador to Applications, and start it from there. On first launch, click Sign in on the welcome view (or the account item in the status bar) and create an account with an email and password. That one account is shared by the editor, the phone app and the hosted agent; there is no API key to paste.

Coming from Cursor or VS Code? Run Celiador: Import Settings from Another Editor from the command palette to bring your settings, keybindings and extensions over; the switching page says what carries over.

VS Code or Cursor, with Celiador Remote

Keep your editor and install the Celiador Remote extension. In Cursor, search for it in the Extensions view; in VS Code, download the VSIX from the latest release and run Extensions: Install from VSIX…. Sign in from its status bar item, then follow the same steps 2 to 6. Runs happen on your machine in a git worktree, the same way; the extension uses the Claude Code CLI you have installed, or downloads it once. Without the Celiador editor you review in the phone app rather than an Agents view.

The cloud (being built)

A run that starts even when your Mac is off, on a server of ours, on a clone of the repository. It is being built and is not live yet. It will need the Celiador GitHub App installed on the repository (step 2), since the push and the pull request happen without your machine. Until then, a labelled issue waits in the queue until a runner of yours is online.

2. Connect GitHub

Two things talk to GitHub on your behalf, and both need a sign-in that most developers already have:

1. The editor pushes the agent's branch and opens the pull request. It uses your git credentials for the push and, for the pull request, either the editor's GitHub sign-in (the Accounts menu in the activity bar) or the GitHub CLI if you have run gh auth login on that Mac. Without either you get a compare link instead of a pull request, so do this before the first run.
2. The Celiador GitHub App hears about labelled issues, comments on them as celiador when a run starts, and notices when the pull request is merged. Install it on the repositories you want agents to work on: github.com/apps/celiador. It asks for issues, pull requests and contents on those repositories, nothing else.

3. Put the queue on your phone

The iPhone app is the review surface: what needs you, what is running, the pull requests waiting, and a merge button. It is distributed through TestFlight. Until the public link is on the download page, ask for an invite from the contact page and mention your Apple ID email. Install it, sign in with the same account as the editor, and you are on the Queue tab.

No phone yet? Everything on the phone is also in the editor's Agents window (Open Agents Window from the Command Palette); you only lose the away-from-desk part.

4. Open the repository and watch it

On the Mac, open the repository folder in Celiador (a normal clone with a GitHub remote) and trust it when asked. The editor tells the queue which repositories it has open. On the phone, tap the eye on the Queue tab and switch the repository on under Watched repos. That link is what routes issues from that repository to your account; one account per repository.

5. Label an issue

Write an issue the way you would brief a colleague for a small change: what to change, where, what not to touch. Create a label named celiador if the repository does not have one, and add it to the issue (or comment @celiador as a collaborator). Within a minute or two:

1. the issue gets a comment from celiador saying it was picked up and on which branch;
2. the run appears under Running on the phone, and in the editor's Agents window, in its own worktree so your working copy is untouched;
3. the agent reads, edits, runs lint and tests, and commits. Routine commands run on their own; a risky one shows up under Needs you with Allow and Deny;
4. when it finishes, the editor pushes the branch and opens a pull request that closes the issue.

Re-adding a label the issue already has does nothing; remove it and add it again to re-run. Up to 3 runs go at once per account; the rest wait in the queue.

6. Review and merge from the phone

The pull request shows under Ready for review. Tap it to read the diff, discard a file, or send a follow-up that starts another turn on the same branch. Swipe the row to Merge: your runner merges it with your GitHub sign-in and the task moves to merged. That is the loop. The account panel keeps count of pull requests merged in the last seven days; it is the one number we steer the product by.

Being built, in the same release cycle: the agent's plan and its questions posted on the issue, so you can answer by replying there; failed CI checks picked up and fixed by the agent; the cost of each run shown on the pull request and on the card; and a self-review section in every pull request. None of that is live yet; when it is, it appears in the changelog first.

When something goes wrong

1. The pull request is a compare link. The editor had no GitHub sign-in; see step 2.
2. Nothing happens after the label. The app is not installed on that repository (step 2) or the repository is not watched from the phone (step 4). A runner must also be online with the folder open: the Celiador editor, or VS Code or Cursor with Celiador Remote.
3. The run failed. The task shows the error on the phone. Our limits are published and enforced per account; the current ones are on the pricing page.
4. Anything else. Open an issue or write to the contact page; a person answers.

The code is open source (Apache-2.0): the editor, the server and the iPhone app. What we do with your code and data is on the privacy page.