Connecting Datadog to AI: know what happened last night without digging
Datadog can be connected to AI. IntraGPT retrieves measurements over a period you name, reads the monitors with their status, searches the logs of a service and can put a release on the Datadog timeline. The question of what went wrong overnight is answered straight from Datadog, instead of somebody opening three dashboards first.
What the assistant can query in Datadog
Measurements over a period
The assistant retrieves the measurement you ask about, across the period you name. Because a period is always required, every figure that comes back says which window it covers.
Monitors and their status
Alert monitors come back with their current state: OK, Warn, Alert or No Data. You can narrow it down by name and by tag, so the monitors of one service or one environment can be looked at on their own.
Search the logs
The assistant searches the logs of a service, within a period, for whatever you ask about. Log volumes are enormous, so it always searches narrowly and never pulls a whole day of lines for the sake of it.
Mark something on the timeline
The assistant can put an occurrence on the Datadog timeline. That is a marker the whole team sees, so it only happens on request and only for something real, such as a release or a maintenance window.
Working with tags instead of names
Datadog hangs everything off tags: service, environment, team, version. The assistant uses those same tags to bound a question, so an answer about staging does not quietly include production numbers.
Where it saves time
Software team
- Situation
- A user reports the application was slow last night, but nobody remembers exactly when.
- What the agent does
- The assistant retrieves the response time for that service across the evening and searches the same period for errors in the logs.
- Result
- A window in which it went wrong plus the first error message, instead of an argument about whose memory is better.
Operations
- Situation
- The day starts and the team wants to know which alerts are still open and which recovered on their own overnight.
- What the agent does
- The assistant retrieves the monitors of the production environment and groups them by status.
- Result
- A readable list in the chat with the state per monitor, so the standup is about the exceptions.
Product management
- Situation
- After a release the product manager wants to know whether the number of errors went up, without interrupting an engineer.
- What the agent does
- The assistant compares the error rate in the period before the release with the period after it and names both windows explicitly.
- Result
- An answer with the periods attached, so the conclusion can be checked.
Infrastructure
- Situation
- Maintenance is planned and afterwards it has to be traceable exactly when it ran.
- What the agent does
- On request the assistant marks the maintenance on the timeline, with the reason and the service it concerns.
- Result
- Next week's graphs show that the dip belongs to the maintenance and not to an incident.
From access to answer
-
1
Connecting from the platform
You grant access from your own Datadog account and it is stored once in IntraGPT. We keep no Datadog password, and you can withdraw that access again in a single step.
-
2
Picking the right region
Datadog runs each customer in its own region. The connector points at the European one by default; if your data sits elsewhere, let us know, otherwise the answer stays empty while the access is perfectly fine.
-
3
Deciding what the assistant may do
Looking along is separate from recording something. If you do not want the assistant marking anything on the timeline, that stays off and the connector only observes. You also decide which teams get to use it.
-
4
Following along in the log
Every use of the connector lands in the IntraGPT log, with what was requested and what came back. That makes it possible to trace afterwards which question retrieved which data.
Why monitoring data deserves care
Metrics look harmless, but log lines regularly carry identifiable data: an email address in an error message, an order number, an address with parameters. That is why this connector is kept narrow.
Where the data lives: Nederland of de EU
Only what the question needs
The connector cannot export the whole log stream. Every search is bounded to a service and a period, so what the assistant gets to see is limited by the question that was asked.
Access stays yours
You grant the access yourself and you withdraw it yourself, and you can limit it to looking along. The connector then cannot change anything in Datadog, whatever the assistant is asked.
Not for every employee
In IntraGPT you decide which roles may use this assistant. One built for the commercial team never sees the monitoring capability at all.
The model runs with us
Answers are composed on our own server in the Netherlands. Log lines and measurements do not travel to an American model vendor and are never used for training.
Frequently asked questions about the Datadog connector
Does IntraGPT work with Datadog?
Yes. Datadog is in the IntraGPT connector catalogue. The assistant retrieves measurements over a period, reads the monitors with their status, searches the logs and can mark an occurrence on the timeline when you ask it to.
Can I connect Datadog to ChatGPT?
ChatGPT has no connection to your Datadog environment, and pasting a log line containing customer data into a public chatbot is unwise. IntraGPT chats the same way, but with a connection tied to access you grant, narrowed per question and recorded every time.
Why do I get no data back while the access is valid?
Usually one of two things. Your Datadog environment sits in a different region than the one the connector points at, or the requested period falls outside how long that measurement is kept. Check the region first, then the period.
What is the difference with the Sentry connector?
Sentry is about individual errors with their trace and the number of affected users. Datadog is about the behaviour of the system as a whole: durations, volumes, monitors and logs. In practice teams use them side by side, and one assistant can query both in a single conversation.
Can the assistant change or mute a monitor?
No. The connector reads the monitors and their status but changes nothing about them. The only thing it can record is an occurrence on the timeline, and that too can stay switched off.
How do I stop the assistant pulling huge amounts of logs?
It is built into the setup: a search is always bounded to a service and a period. The assistant also searches narrowly rather than broadly. If a question still gets interpreted too widely, we tighten that assistant's instruction.
What does a Datadog connector cost?
There is no separate price tag. Connecting is part of the track we go through together. What drives the work is how many sources are added, whether the assistant may record anything and which teams get to use it. We make that concrete in a thirty minute conversation.
How long does setup take?
Once the access is arranged and the region is known, switching it on is a matter of minutes. The lead time usually sits in agreeing internally who grants that access and what rights hang off it.