Connecting Sentry to AI: knowing which error hits the most users
Sentry can be connected to AI. Sentry collects the errors that occur in production; IntraGPT retrieves them inside the project you point it at, sticks to the period and the question you name, and shows how often an error occurred and how many users it hit. Marking an issue as resolved is possible too, but only when you ask for it.
What the agent pulls from Sentry
Retrieve open errors
The agent fetches the errors of the project that have not been resolved, over the period you name and in a list you can take in. If there is more, it asks for the next part rather than putting everything on the table at once.
Weighing by impact
For each issue the agent looks at how often it occurred and how many users it hit, not only at the wording of the error message. Sorting by frequency shows what hurts the most, rather than what happened to arrive last.
The trend over a period
Over a period you choose yourself, the agent shows how the numbers developed, per day or more finely. That makes it visible whether a peak coincided with a release or with a busy day.
Mark an error as resolved
The agent can set an issue to resolved and assign it to someone. It does that only when the cause is really gone, because otherwise the error returns as a regression and the list stops meaning anything.
One project per connector
The connector is fixed to the organisation and the project in Sentry that you point it at. Figures from another project therefore never leak into an answer.
Moments where this saves time
E-commerce
- Situation
- After a release something seems wrong with checkout, but the reports reaching customer service are vague.
- What the agent does
- The agent retrieves the open errors of the past day and ranks them by the number of users affected.
- Result
- Within a few minutes it is clear whether this is one visitor on an old browser or a real outage.
Government
- Situation
- A digital counter has to be available and an administrator wants to know how the week looked without opening a dashboard.
- What the agent does
- The agent shows the trend over that week, with the numbers per day.
- Result
- A picture of peaks and quiet days, with the option to dig into a day that stands out.
Customer service
- Situation
- A customer reports an error message and the service desk wants to know whether this is already known.
- What the agent does
- The agent searches on a term from the report and checks whether an open issue matches it.
- Result
- The customer hears whether it is known, and the same fault is not filed for the fourth time.
Industry and manufacturing
- Situation
- A fix for a fault in a control portal has been rolled out and the issue can be closed.
- What the agent does
- On request the agent sets the issue to resolved and assigns it to whoever picked it up.
- Result
- The list stays clean, and if the error does come back it shows up as a regression.
Connected in four steps
-
1
Connecting to your Sentry
You pick Sentry in IntraGPT and connect it with an access token you issue yourself and can withdraw again.
-
2
Bounding it to one project
You point out the organisation and the project the questions are about. If several projects are running, each gets its own connector so the figures do not get mixed up.
-
3
Deciding what may be closed
Reading along and viewing the trend are separate from marking an issue. Leave that last one out and the connector stays purely informative.
-
4
Roles and recording
You decide which roles may use this agent, usually development and operations. Every action is recorded with the moment and the outcome.
Personal data in error reports
An error report is not a neutral technical line: it can easily contain the email address of the affected user or a URL with parameters in it. The connector takes that into account.
Where the data lives: Nederland of de EU
What a stack trace can hold
Sentry records who ran into the error and on which page. That may be an email address or a URL carrying details of that user. The agent therefore retrieves only the details belonging to the question and builds no user profile out of error reports.
A token with the right reach
You issue the access token and decide which organisation and projects it reaches. A token that may only read cannot mark anything, however the question in the chat is phrased.
Resolved means resolved
The agent marks an issue only on request. If the cause is not really gone, the error returns as a regression and the list has lost value. Tidying for the sake of tidying is not part of this.
Error data stays inside
The model runs on our own server in the Netherlands. Stack traces and user details from Sentry are not sent to an American model provider and are not used to train a model.
Questions about Sentry and AI
Does IntraGPT work with Sentry?
Yes. Sentry is in the IntraGPT connector catalogue. The agent retrieves the errors of a project, shows the trend over a period and can set an issue to resolved, all inside the project you point it at.
Can I connect Sentry to ChatGPT?
Not without handing your error data and an access token to a public service, and a stack trace can hold details of real users. IntraGPT offers the same way of asking, but the connector is tied to one project and every action is recorded.
Why does a resolved error keep coming back?
Because Sentry reopens an issue that occurs again after being marked, as a regression. That is exactly the intention: resolving is a statement that the cause is gone, not a way to make the list shorter. If an issue returns, the fix was not in yet or it did not cover every case.
Does the agent see personal data in a stack trace?
It can, because an error report sometimes contains the email address of the affected user or a URL carrying details about them. The connector therefore retrieves only the details belonging to the question and does not search wider than needed. If you want certain data never to appear, you hold it back in Sentry itself when an error is recorded.
How does the agent know which project to look at?
While connecting you point out the organisation and the project in Sentry. If you work with several projects, a frontend and a backend for instance, you set those up separately so the figures do not run into each other.
Can the agent fix an error itself?
No. It reads the errors, weighs them by count and users affected and can mark or assign one. It does not change code. For that the work goes to GitHub or GitLab, where an issue or a merge request belongs.
What does a Sentry connector cost?
The connector belongs to the project we run together. What matters is the number of projects you connect, the rights on the token and whether the agent may close things as well. We lay it out in a thirty minute call.
How long does setup take?
With an access token and a designated project it is a matter of minutes. More time goes into agreeing who may use the agent and whether it may close issues or only look.