Grok Ship 2026: Firstmate Files It, You Merge?
Key takeaways
- You talk to Firstmate. It files the task and hands the code to a Bot for that project.
- A scout comes back as a report. A ship comes back as a pull request after a fresh look at the branch.
- Grok Ship does not merge unless you say so, and it does not merge while checks are red.
- The template is MIT, by Kun Chen. A new plan that includes the Bot is Premium at $30 a month.
You keep one chat open, with a Bot named Firstmate. It writes the task down and hands the code to a Bot for that project. Ask for a diagnosis and you get a report. Approve a fix and you get a pull request. You merge it.
| Work | Firstmate and Grok Ship | Cursor or Claude Code |
|---|---|---|
| You want a diagnosis, not a code change | Scout. The report stays on the shared computer. No pull request | You open the repo and read it yourself |
| You already approved a fix | Ship. A cloud agent, a fresh review of the branch, then a pull request | The editor all day is Cursor. A long terminal session is Claude Code |
| Checks are red | Grok Ship does not merge that | A red check is still yours to clear in either tool |
Talk only to Firstmate
Firstmate is the only chat you keep. The other Bots each own one project. Firstmate messages that Bot, marks the ask with a task number, and waits for the reply on that same number. If nothing changed, it still reports back.
A login stays with the Bot that needs it. The others cannot use that login. That Bot asks, and you hand the secret over on the card the app shows for a secret. Do not paste it into the chat. Firstmate’s instructions say the secret stays with that Bot, and that Firstmate never starts a Cursor cloud agent. The project Bot does.
The code runs on a cloud agent, and you still merge
Nothing here runs on your laptop. The task list sits in a database on the shared Grok Bot computer, which is also where a scout report lands. The code change runs on a Cursor cloud agent that ends when the task ends.
A scout is a diagnosis, a plan, or an audit. What you get is a report, and no pull request. A ship is a change you already approved. If that scout should become a code change, it updates the same task instead of opening a second one.
The cloud agent pushes a branch, and a new review reads that branch before any pull request opens. The Grok Ship README says those ships never merge without you, and never while checks are red.
A triage Bot is extra. You can ask Firstmate to add one, and that Bot is not in the first install. It may merge a small correction only when checks are green, the project’s written goal still matches, the review leaves no question open, and the change is not the product’s usual behavior and not a security fix.
It does not assume the code is on GitHub. GitHub, GitLab, Bitbucket, or Cursor Origin can be the host.
Install Grok Ship from the template
The README opens with a warning: this repository has been replaced. Install from the shared template that warning names. Keep the README for what a scout is, what a ship is, and who merges. An older quick start told a Bot to follow a markdown file in the repo. Use the template.
Kun Chen reported hundreds of issues and pull requests in a day
Kun Chen wrote that one day this got through hundreds of issues and pull requests on his repos. The line is in the note he posted. Then decide whether your next task should come back as a report or as a pull request you merge.
Cursor or Claude Code when you want the editor or the terminal
A morning of account research, or a digest, is a different job. Grok Bot use cases starts there, and that first run stays a list you review. It does not file a scout or open a pull request.
If you live in the editor all day, or you want a terminal agent on a Claude plan, weigh the editor against the terminal in Cursor vs Claude Code before you add Firstmate. If you already know the work stays in the terminal and you are choosing Claude against Codex, use Claude Code vs Codex.
Either way, the Bot plan is the same. As of October 2026, a new plan that includes Grok Bot is Premium at $30 a month. Plus at $8 does not include it. Cloud Agents, which this template uses for the code change, start at Premium. If an older plan still includes the Bot because you never switched, that exception is on the Grok Bot page.
Pick Firstmate, or stay in the editor
I’d pick Firstmate when I want the task written down, and I want either a report or a pull request I still merge. I’d stay in the editor, or in the terminal, when I do not want another Bot between me and the change.
How I checked this
This page is who you talk to in Grok Ship, and who merges. I used the Grok Ship README, Firstmate’s instructions on GitHub, and Kun Chen’s note. Plan prices are from Cursor’s pricing page. I pick by whether you want Firstmate to file the task, then by whether you still merge.
Products in this guide
- Grok Bot The cloud computer Grok Ship runs on. You talk to Firstmate. Code changes run on a Cursor cloud agent. You merge. Visit site
- Cursor The editor, and the cloud agents this template uses when the task is a code change. Visit site
- Claude Code Anthropic's terminal agent, when you do not want a Bot filing the task. Visit site
FAQ
Do I only talk to Firstmate?
Yes. Firstmate files the work, hands it to the project Bot with a task number, and brings the result back. It does not start the Cursor cloud agent. The project Bot does.
Does Grok Ship merge pull requests for me?
A normal ship waits for you, and it will not merge while checks are red. A triage Bot you add yourself may merge a small fix when checks are green, the project’s written goal still matches, the review leaves no question open, and the change is not the product’s usual behavior and not a security fix.
Is Grok Ship an xAI product?
No. Kun Chen published it under MIT. The README sends you to a shared Grok Bot template. You still need a plan that includes Grok Bot.
