Skip to content
Development Working integration

Connecting GitLab to AI: from merge request to pipeline in one conversation

GitLab can be connected to AI. IntraGPT works inside the project you point it at, reads issues and merge requests there, creates an issue on request and can start a pipeline on a branch you name. If GitLab runs under your own management, the same connector works there as well.

A release rarely stalls on the code itself. It stalls on the question of which merge requests are still open, which issue belonged to that one again, and whether the pipeline on that branch was green. Three questions, three tabs, and usually someone diving in while they were really busy with something else. The GitLab connector fetches that information where the question is asked. The agent works inside one project that you point it at and never looks into another. It searches issues with the boundary you supply, for instance only what is open or only a certain label, shows which merge requests are running and with whom they sit, and on request files an issue with a description, labels, an assignee and a target date. There is also an action that starts a pipeline. That is no rehearsal: such a run builds and tests, and depending on how you set things up it can push something to production. So the agent starts one only after an explicit request, and names the branch it will use before it begins.

What the connector does in a GitLab project

Search the project issues

The agent searches issues within the project and sticks to what you ask for: only open ones, a certain label, a milestone or a particular colleague. Filtering is not a luxury, because without a boundary an active project returns more lines than anyone reads.

Create an issue with a target date

On request the agent files an issue with a title, a description, labels, an assignee, a milestone and a date by which it should be done. That puts a report into GitLab in full, instead of half in a chat and half in somebody's head.

Review merge requests

The agent shows which merge requests are open, which are yours and which are waiting on someone else. The answer stays short enough to read, and you can drill into any one of them.

Start a pipeline

The agent can start a pipeline on a branch you name, with variables for that environment if needed. That is a real run that takes time and that can build, test and deploy, so it comes back to you explicitly first.

Also on a self-managed GitLab

If GitLab runs on your own infrastructure or only inside your network, you point the connector there. Same possibilities, same limits, only a different place where GitLab lives.

Examples from practice

E-commerce

Situation
Just before a promotion week a webshop team wants to know whether any merge requests touching checkout are still open.
What the agent does
The agent retrieves the open merge requests carrying that label and lists their titles.
Result
The team decides on the basis of a list instead of a gut feeling.

Industry and manufacturing

Situation
A fault on the shop floor has to land as an issue in the right project, with a date by which it must be solved.
What the agent does
The agent files the issue with a description, labels, an assigned colleague and that date attached.
Result
The report sits complete in GitLab without the reporter having to log in.

Release management

Situation
After an approved merge the acceptance environment has to be rebuilt.
What the agent does
On explicit request the agent starts a pipeline on the named branch, with the variables belonging to that environment.
Result
The build runs, and the conversation records which branch was started and why.

Support

Situation
A customer asks whether a fault reported earlier is being worked on.
What the agent does
The agent finds the issue by label and status and checks whether a merge request belongs to it.
Result
An answer with the issue number attached, without disturbing a developer.

How it is made ready

  1. 1

    Connect the project

    You pick GitLab in IntraGPT and connect the project with an access token you issue yourself. If GitLab runs under your own management, you point at that environment instead of the public one.

  2. 2

    Decide what is allowed

    Reading, creating an issue and starting a pipeline are released separately. Many teams begin with reading along only and extend that later.

  3. 3

    Release pipelines on purpose

    Because a pipeline sets something in motion, that is a separate choice. Leave it off and people can ask for it in the chat all they like: nothing happens.

  4. 4

    Visibility and recording

    You decide which roles see this agent. Every action, a started pipeline included, is recorded with the moment, the branch and the outcome.

Connected securely

CI rights and what the agent may and may not do

A CI system can set things in motion that go further than a line of text in a ticket. The connector is built with that in mind.

Where the data lives: Nederland of de EU

The token sets the boundary

You issue the access token and decide what it may do. Give it read rights only and starting a pipeline is plainly impossible, however the question in the chat is phrased.

A pipeline never starts by itself

A run begins only after an explicit request, and the agent names the branch it will start on before doing so. Building because something happened to come up in a conversation does not occur.

Not every employee sees this agent

You tie the GitLab agent to the roles that have something to do with it, usually development and operations. For the rest of the organisation it does not exist.

Own server, no training

The language model runs on our own server in the Netherlands. Issue text and branch names from your project are not sent to an American model provider and are never used for training.

Questions about GitLab and IntraGPT

Does IntraGPT work with GitLab?

Yes. GitLab is in the connector catalogue. The agent searches issues, creates one, reviews merge requests and can start a pipeline, all inside the project you point it at.

Does it work with a GitLab we host ourselves?

Yes. While connecting you indicate where your GitLab runs, and that may be an environment reachable only inside your own network. The possibilities stay the same and the rights still hang on the token you issue.

Can I connect GitLab to ChatGPT?

ChatGPT has no connection to your GitLab, and certainly not to an instance that is reachable internally only. IntraGPT offers the same way of asking, but with a connector that stays inside one project, with a token whose reach you determine and with every action recorded.

Can the agent put something into production?

Only indirectly, and only if your pipelines do that. Starting a pipeline sets a run in motion on a branch; what that run then performs is something you configured yourself. If it deploys to production, that happens when the agent starts it too. Which is why the agent asks explicitly and names the branch.

Can the agent see what is in the code?

No. The connector is about issues, merge requests and pipelines, not about the contents of files in the repository. The agent reads what has been written down about the work, not the source code itself.

How does this differ from the GitHub connector?

The reading side looks similar, but GitLab adds starting a pipeline and works with merge requests rather than pull requests. In exchange, the GitHub connector can comment under an issue. If you use both platforms the connectors can sit side by side; each is tied to its own project or repository.

What does a GitLab connector cost?

Connecting is part of the project we run together. The size depends on the number of projects, the rights on the token and whether the agent may write or build as well. We lay that out in a thirty minute call.

How long does setup take?

With a token and a designated project at hand the connector is live within minutes. The lead time sits in the internal weighing of how much that token may do and who is allowed to have a pipeline started.

For which sectors

Other connectors

See all connectors

Curious what this connector would deliver for you?

Book a free thirty minute AI call. We look at your systems, the permissions around them and the first use case that saves time or money.

Book a free AI consultation