Once-sceptical Grafana says AI helps pull observability dashboards from "thin air"
Observability firm puts AI, APIs on agenda
grafana
Grafana tied its future to APIs and and AIs this week at its customer get together in Seattle.
The observability specialist unwrapped the latest release of its core platform, Grafana 12, delivering what it dubbed “observability as code” and threading agentic AI into the platform.
It has launched an LLM-based Grafana Assistant in its Grafana Cloud product, essentially a chatbot to connect users to observability data, answering natural language questions and providing “relevant suggestions”, fault finding and even generating dashboards.
The company said the potential use cases were “limitless” but flagged examples such as asking questions about observability data, making bulk changes, and creating new dashboards through natural language descriptions.
A live demonstration of the technology saw the assistant identify data connection problems in infrastructure, and create a dashboard “out of thin air”.
Principal software engineer Mat Ryer said “Sometimes it has more knowledge than you in certain places, and sometimes you're one that has the knowledge. So you work together in this integrated way.”
Cofounder of Grafana Labs, Torkel Ödegaard, said the company had used Anthropic models to underpin its AI strategy.
It had been hesitant about overselling the technology, he said, but “We're still kind of a bit blown away [with] what we've been able to do in a very short, kind of short amount of time, exploring the possibilities there.”
Speaking to The Stack last year, Ödegaard said the Grafana community wasn’t necessarily clamouring for AI features.
This week, Ödegaard said its AI strategy lay in “Figuring out how to leverage these tools to make every person that uses Grafana an observability expert, to level them up in knowing how to troubleshoot, how to sort of perform data analysis.”
Asked what this implied for actual “experts”, Ödegaard said, “hopefully it means faster trouble shooting…not just having to reach out to or wake up the expert.”
Grafana CTO Tom Wilkie said that earlier experiments with AI had been disappointing, but that it had a breakthrough around six months ago, as models improved. “In particular, they have this have this instruction following post training step…this is what really made Anthropic special.”
At the same time, Grafana has launched an "App Platform”, which it described as the “backbone of the observability as code strategy”.
This essentially means a set of “consistent, versioned APIs for managing Grafana resources” and a set of tools for building custom apps on Grafana.
Ödegaard said this has involved a major architectural exercise for Grafana.
“Many people have different ideas about what infrastructure as code is, how it should be built. So it’s just taken us a while to get to a point where we agree on how to build this and what the fundamentals should be.”
He said the new approach was built around the same versioned API approach as Kubernetes. “It's going to sort of set us up for success in the sense that the API machinery is going to be very familiar, and we're going to be compatible with a lot of the Kubernetes infrastructure automation.”
It has also launched a new JSON schema for dashboards, that decouples general settings and content. These will supply more flexible dashboard creation and more customization options.
Dashboards can now be synced to a GitHub repo, with new “as code tools” which will allow easier integration into pipelines or GitHUb actions.
Users also get 15 new data sources, include DynamoDB, CosmoDB, Atlassian and PagerDuty.