Your AI ships the code. Octopad closes the loop.
When your AI does the work, your plan can't be the one thing still updated by hand.
Your team works through AI now, and not just for suggestions. Your coding agent picks up a task, writes the change, opens the pull request, and waits for a person to look. That part works.
Here is the part that does not. The task that pull request finished still says "in progress" three days after it merged. Nobody moved it. The plan and the code have quietly drifted apart, and the more your AI ships, the wider the gap gets.
It is not a discipline problem. It is a structure problem. Your AI can open the pull request, but it cannot keep your plan honest, because the plan lives in a tool built for people to update by hand. The AI can talk about the work. It cannot reliably keep the record of it.
The tax you are already paying
So someone does it by hand. At the end of the day you open GitHub, see what merged, then open Jira or Linear or Notion and move the matching cards to done. Five minutes here, ten there, and it is the first thing that slips when you are busy. When it slips, the plan rots. Standup turns into "wait, is that shipped?" The board becomes a thing nobody trusts, which is worse than no board at all.
And it keeps happening for one reason: "done" lives in the wrong place. Today it is a card someone has to remember to move, not the moment the work actually shipped. Fix where "done" comes from and the tax goes away.
"Done" is a merge, not a message
Here is the reframe. Done is not when the agent says it is done. An agent is a confident narrator: it will tell you the task is finished, formatted and sure, whether or not anything landed. Done is when a person reviews the pull request and merges it. The merge is the ground truth for shipped.
So the plan should read from the merge, not from the agent's word and not from someone's memory. A plan that cannot see the merge is guessing.
Where Octopad fits
Octopad is your AIs' back-office. It organizes your company's work and knowledge so the AI each person already uses can run the work from one source of truth. It does not run the work. Your AI does that. Octopad keeps the record true.
For GitHub, the split is simple: a task in Octopad can be tied to the pull request that finishes it, and Octopad keeps the two in step. The link gets made one of two ways.
- Your coding agent adds one line to the pull request description, a small marker naming the task. The link is made as the agent opens the PR. Nothing runs in your repository, and there is nothing to install. The marker is just text in the description.
- Or you point a task at a pull request directly, with no marker involved. Your AI can do it for you in Octopad, or you can do it yourself. It covers a PR a person opened, or any that went up without the marker.
From there, nothing changes about how you ship. A person reviews the pull request and merges it, exactly as before. The moment it merges, Octopad moves the task to done and leaves a note on it linking the pull request. If there is discussion on GitHub, or an issue closes, that lands on the task too. The board stays true without anyone tending it.
What it looks like
Say you hand your agent a task from the plan: add a rate limit to the login route. It writes the change and opens a pull request with the one-line marker in the description. You get on with your day.
A colleague reviews the PR, asks for one change, gets it, and merges. Before you have refreshed the board, the task is done, with a comment on it that links straight to the merged pull request and names who merged it. Nobody moved a card. The plan matched the code the second the code landed.
Issues, both directions
Pull requests are where code lands, but issues are where a lot of work still gets tracked, and not everyone on your team lives in the plan. Octopad meets that too.
You can turn a task into a GitHub issue in one step, so the work shows up for the people, and agents, who work out of GitHub, linked back to the task from the moment it is created. Or you can link a task to an issue that already exists, and pull that work into the plan. Either way, once they are linked, GitHub keeps the task honest: when the issue closes, the task closes in Octopad, and comments on the issue show up on the task. One piece of work, one truth, whichever surface it started on.
What Octopad does not do
Octopad does not write your code. It does not run your agent. It does not test, review, or merge your pull requests. A person does that, and that is the point. The merge is a human call about what is good enough to ship, and it stays a human call.
| Your AI runs | Octopad organizes | You decide |
|---|---|---|
| Picks up the task, writes the change, opens the PR | Ties the task to the pull request or issue, keeps the plan current | Review, and merge or close |
Humans set the direction and hold the merge. The AI does the structured work underneath. Octopad keeps the plan true to what actually shipped.
It gets better with a team
On your own, this saves you the end-of-day reconciliation. With a team, it does something you cannot do by hand.
Every engineer works through their own AI, in their own chat, on their own tasks. What they share is the plan, and the plan stays current from one place: the merges. When one person's agent ships a change and another person merges it, everyone's next session opens onto a true picture of what is done and what is left, the same context across every AI on the team, with no sync meeting to get there.
The more a team leans on AI to do the work, the more it needs one source of truth the work reports back into. That is the job Octopad is built for.
Getting started
Connect your repository to Octopad once, add the marker convention to your project's instructions so your agent includes it, and the next pull request that merges closes its task. It runs in minutes, and it works with whatever coding AI your team already uses.
Start free and link your first repository.
Common questions
Do I have to install anything in my repository?
No. Your agent adds one line to the pull request description. Nothing runs inside your repo.
Which AI tools does this work with?
Any coding AI that can open a pull request. Claude Code, Cursor, and the rest, because the link lives in the pull request description, not in a special connection to one vendor.
Does Octopad merge or close things for me?
No. A person reviews and merges, and a person closes an issue, the same as today. Octopad updates the plan after. It does not make the call about what ships.
Does this replace GitHub, or my tracker?
No. GitHub stays the home for your code and your issues. Octopad is the plan your AIs run the work from, kept true by what merges and closes. If you track work somewhere your AI cannot read or update, that is the gap this closes.
What about our data?
Your workspace content is encrypted in transit and at rest, access is scoped to the workspace and the roles you set, and nothing is ever used to train AI models. Octopad connects to your repository with access you grant and can revoke.