What's new from Grafana Labs

Assistant Investigations is now generally available. It’s Grafana Assistant’s way of running longer, deeper investigations for you: point it at an alert or incident, and it explores your metrics, logs, traces, and profiles, builds hypotheses as it learns more, and produces a structured report with findings, supporting evidence, and recommended next steps.

With Agentic Testing, describe a user flow in plain language, like “log in, add an item, check out,” and an agent runs it in a real browser on Grafana Cloud: locating elements, adapting to UI changes, navigating unexpected states, and verifying the result looks right. No brittle selectors to maintain, no scripts to babysit.

Watchers are a new Grafana Assistant capability that add an agentic option alongside your existing alerts. They’re built for the cases that are hard to express as a traditional rule: sometimes you can’t spell out every way something can go “wrong,” but you know exactly what “right” looks like. Ask the Assistant to set up a Watcher, describe what healthy looks like, and it keeps watch for when reality starts to drift.

Maintenance windows are now available for Grafana SLOs. Use maintenance windows to schedule planned work for one or more service level objectives (SLOs). During an active maintenance window, Grafana SLO pauses error budget consumption and burn rate alerting for the selected SLOs.

We are reducing the number of deprecated and outdated plugins running in Grafana Cloud. You can no longer install plugins that have been deprecated for more than 6 months. Existing installations are not affected, and the error message points to the deprecation notice in the plugin catalog, which usually names an alternative.

As part of AI Week, we’re shipping a restructured information architecture (IA) for Grafana Cloud docs. The new navigation makes it easier to find the right docs by grouping content around how you work in Grafana Cloud, with a clear home for AI capabilities.

This change reflects how the platform has actually grown.

The new structure groups everything into four sections:

• Platform - cross-cutting capabilities that support Grafana Cloud’s product suite
• Observe and act - the task-oriented home for monitoring, alerting, incident response, testing, and data connections, where product names (Alerting, IRM, SLO) now live on landing pages instead of in the navigation.
• AI tools - infrastructure-level AI capabilities (MCP server, Dynamic alerting, gcx, the LLM plugin) that enhance workflows rather than standing alone.
• Learn and build - Grafana Learn, get started, conceptual content, dashboard-building, and developer resources.

Selecting an OpenSearch index just got easier with the new Index Picker.

Finding and setting the right index in the Grafana OpenSearch plugin no longer means typing index names from memory. The query editor now includes a Select index button that opens a searchable browser of the indices available on your data source. Each index is shown alongside its document count and cluster health, so you can pick the right one at a glance. Search by substring, use * and ? for wildcards, or wrap your input in /…/ for a full regex match to quickly narrow down large clusters.

Give your AI agents instant, zero-infrastructure access to your entire observability ecosystem, data, alerts, and incidents, using a single managed endpoint. Not to be mistaken with the OSS MCP server, which you install, host, and maintain yourself, the Cloud MCP server is fully managed by Grafana Labs, so you can supercharge agentic workflows without deploying or running anything of your own.

gcx is now generally available for Grafana Cloud and Grafana OSS/Enterprise v12 or later. It’s the CLI that gives you and your AI agent structured access to Grafana and Grafana Cloud from the terminal — dashboards, folders, alert rules, and data sources alongside Synthetic Monitoring, k6, Fleet Management, Incidents, Adaptive Telemetry, and more. As the evolution of grafanactl, it brings a previously fragmented experience into one tool and integrates directly with Grafana Assistant.

We’re excited to announce the private preview of Grafana Assistant powered by OpenAI models. Anthropic models have been the backbone of Grafana Assistant so far. Now you can try the Assistant experience powered by OpenAI instead: when enabled, all requests are routed to OpenAI models rather than Anthropic ones.

The fastest way yet to see what Grafana Assistant can do. Head to play.grafana.org and start a conversation right there in our public sandbox. No install, no account, no setup — just you and the Assistant on real Grafana data.

Workspace, the agent-native interface for Grafana Cloud, is now generally available in Grafana Cloud. Instead of hopping between tabs to pull a query here, a dashboard there, and an incident timeline somewhere else, you and the Assistant work side by side in a single, full-page view: building reports, correlating signals, and running everyday observability tasks without ever leaving the window.

Understand who is using Grafana Assistant and how AI token usage contributes to your costs directly in the Cost Management and Billing App.

We’re excited to introduce Grafana Assistant Usage Attribution, giving administrators visibility into Assistant token consumption across their organization. Available  in the Usage view of the Cost Management and Billing App, it surfaces alongside your existing Grafana Cloud usage data. You can see token usage by individual users and service accounts, distinguish between included and additional token usage, and manage Assistant usage alongside your other Grafana Cloud products from a single experience.

As you write a performance test in Grafana Cloud k6, the script editor can flag hardcoded API keys, tokens, and passwords before you run the test. Those values are stored in plaintext in the script, so leaving them there puts credentials at risk when you share the file or check it into version control.

Cost-effectively get the right representative mix of traces with volumetric policies!

Now Generally Available, volumetric policies intelligently balance cost, service representation, and value to automagically give you the right traces. We’ve bolstered volumetric policies under the hood to be even more cost predictable.