Connecting GitHub to AI: knowing what is open without opening the repo
GitHub can be connected to AI. IntraGPT works inside the repository you point it at, looks up issues and pull requests there, and can create an issue or comment on one when you ask. Questions such as what is still open for this milestone or which pull requests are waiting for review get answered in the conversation, without anyone opening the repository.
What the assistant can do in a repository
Find the issues that matter
The assistant looks up issues in the repository and keeps to the boundaries in your question: only open ones, only a certain label, only what belongs to a milestone or what is assigned to someone. A full issue list is too long to work with in practice, so there is always a limit attached.
Create an issue
On request the assistant files a new report in the repository, with a title, an explanation, labels and an assignee. Because the whole team sees that report, it writes a concrete title and puts in the body what goes wrong, what was expected and how to reproduce it.
Review open pull requests
Running pull requests come back with the branch they come from and the branch they are headed for. You can ask for only the open ones or for what was updated most recently, so it is clear what is still waiting for review before a release goes out.
Comment on an issue
The assistant can post a comment under an existing issue. That comment is readable by everyone with access to the repository, which for a public repository means the entire world, and it stays in the history.
One repository per connector
The connector is pinned to the repository you point it at. The assistant stays inside it and cannot wander off to another repository of the same organisation.
Where this saves the team time
Industry and manufacturing
- Situation
- The software crew at a machine builder gets fault reports through the service desk, but those never reach the repository.
- What the agent does
- The assistant turns a report into an issue with a reproducible description, the right labels and the colleague involved attached.
- Result
- One place where the work sits, and the service desk does not need a GitHub account of its own.
Product development
- Situation
- Before the sprint changeover the team lead wants to know what is still open on the current milestone.
- What the agent does
- The assistant looks up the open issues on that milestone and adds who they are assigned to.
- Result
- A list inside the conversation instead of a screenshot from the project board.
Government
- Situation
- A municipality has an application built in an open repository and must be able to show what happened to a reported fault.
- What the agent does
- The assistant finds the issue by label and status and pulls in the pull requests that were merged into the same branch.
- Result
- The path from report to change can be retraced in a few sentences, with the numbers attached.
Release management
- Situation
- Just before a rollout it is unclear which pull requests have not been reviewed yet.
- What the agent does
- The assistant retrieves the open pull requests aimed at the release branch, most recently updated first.
- Result
- The release manager knows within a minute what is still blocking.
From connecting to the first question
-
1
Connect through the platform
In IntraGPT you pick GitHub from the catalogue and connect the repository with an access token you issue yourself. No employee password is stored.
-
2
Decide what is allowed
You choose what the connector may do: read along only, or also create an issue and comment. Whatever is not released simply does not exist for the assistant.
-
3
Decide who may use it
You point out the roles that may work with the GitHub assistant. For the rest of the organisation it does not appear in the list of assistants.
-
4
What gets recorded
Every action is recorded with the moment and the outcome. A created issue or a posted comment can therefore be traced back to the request that preceded it.
Who reads along in the repository
A repository holds more than code: customer reports, internal discussions and sometimes a piece of configuration that has no business outside the team.
Where the data lives: Nederland of de EU
A token you manage yourself
The connector works with an access token you issue and can withdraw at any moment. Give it read rights only and nothing in GitHub changes, whatever gets asked in a conversation.
Public really is public
A comment or a new issue in a public repository is readable by anyone on the internet and stays in the history. That is why the assistant never writes on its own initiative and keeps internal detail out of a public report.
Traceable who asked for what
Created issues and posted comments can be found back with their moment and the request behind them. An administrator can therefore check where a report came from and who asked for it.
The model runs with you
IntraGPT uses our own server in the Netherlands. Text from issues and pull requests is not sent to an American model provider and is not used to train a model.
Frequently asked questions about GitHub and AI
Can I connect GitHub to ChatGPT?
ChatGPT has no built-in connection to your repository, and pasting an access token into a public chat is not a good idea. IntraGPT works like a chat, but with a connector tied to one repository, with a token whose rights you decide and with every action recorded.
Does IntraGPT work with GitHub?
Yes. GitHub is in the IntraGPT connector catalogue. The assistant looks up issues, creates one, reviews pull requests and can comment on an issue, all inside the repository you point it at.
Why does the connector not see our repository?
Almost always because the token has no access to that repository. For a private repository GitHub acts as if it does not exist rather than reporting missing rights, which makes it look like a typing mistake in the name while it is really about access. So check whether the token belongs to this organisation and this repository and whether it is still valid. A misspelt name is the second common cause.
What happens when the assistant comments in a public repository?
That comment is immediately readable by anyone on the internet, under the account used for the connection, and it never leaves the history. This is why the assistant comments only after an explicit request and shows the text first. If you do not want that risk at all, simply do not release commenting.
Can the assistant search several repositories at once?
No. One connector works inside one repository. If you need more, you set them up separately, each with its own token and its own rights. That also keeps it clear which team can reach which code.
Can the assistant merge or close a pull request?
No. The connector reads pull requests and can comment on an issue, but merging, closing or changing labels on a pull request is not included. What ends up in the code stays a decision made by a person in GitHub itself.
How long does setup take?
Once the token exists and the repository is known, the connector is ready in minutes. The time usually goes into the internal agreement on what rights that token gets and whether the team wants the assistant to write as well.
What does a GitHub connector cost?
Connecting is part of the project we run together. What drives the work is the number of sources you want to open up, the rights that go with them and whether writing is involved. We make that concrete in a thirty minute call.