Use AI and automation on Grafana Labs

Investigate in plain language

Wed, 17 Jun 2026 12:54:46 -0500

Grafana Assistant

Grafana Assistant in Grafana Cloud adds the higher-order, agentic features that need the full backend:

Infrastructure memory: Assistant builds and retains context about your environment over time.
Hosted Cloud MCP connections: Zero-install, OAuth 2.1 MCP. Self-managed users run the local OSS MCP server with a manual service account token instead.

CLI auth tokens, anonymous access, and Slack integration: Cloud-native access and collaboration paths

What keeps you up at night?

Role / Worries	What you get with Grafana Assistant in Grafana Cloud
SRE, On-call Engineer Getting paged with no context Spending 20 minutes writing queries just to understand what’s happening Context-switching between tools while the clock runs	Infrastructure memory pre-maps your services and dependencies; no time is lost rebuilding context when a page fires @Grafana in Slack surfaces dashboards and answers without leaving the incident channel
Developer Never sure you’re querying the right data source No idea what your service depends on upstream	Infrastructure memory already knows your environment, dependencies, and which metrics matter
Senior Engineer, Resident Grafana Expert Receiving all escalations for dashboards and queries Spending more time unblocking others than doing your own work	Skills reference your actual dashboards and metric names; team gets environment-specific guidance, not generic advice
Engineering Manager, VP Engineering Best engineers stuck on toil Knowledge concentrated in one or two people A growing stack most of the team can’t use	Infrastructure memory gives any on-call engineer immediate context on services and golden signals, not just seniors
Platform Engineer, DevOps Runbooks that go stale A gap between what the stack can do and what the team knows to do with it	Skills with MCP auto-approve pre-authorize tool calls to GitHub, Jira, and Slack in a remediation flow; no manual approval mid-incident
Grafana, Platform Admin Controlling who can use AI features Preventing Skills from triggering actions they shouldn’t Managing the Cloud vs self-managed feature gap	Assistant-specific RBAC roles control feature access at a granular level MCP auto-approve requires each Skill to declare pre-authorized tools; nothing runs without deliberate configuration Self-managed connects with one click; Cloud-only features are hidden in its UI automatically
Security, Compliance Officer AI models trained on sensitive production telemetry Prompts leaving your control No audit trail for what the AI did during an incident	Requests always route through Grafana Cloud, never directly to AI providers, and only minimal prompt context is sent Assistant operates within the user’s RBAC scope; it can’t see data the user can’t see Full tool use and reasoning history in every conversation gives a complete audit trail

Tackle trends and anomalies with machine learning

Wed, 17 Jun 2026 12:08:02 -0500

Get ahead of problems before users notice

Fixed thresholds have real limitations.

Limitation	How it affects you
Reactive by design: Alert fires only after the threshold is crossed.	Reacting to capacity issues instead of preventing them
Can’t express how one instance differs from the rest	Outliers hide in aggregate data
No trend projection	Guessing at capacity planning instead of forecasting it

With Grafana Cloud Machine Learning, you can catch what threshold alerting can’t:

An instance trending toward trouble
A single node quietly diverging from the rest of the cluster

Grafana Cloud machine learning offers forecasting and outlier detection.

Real results from real teams

MediaKind

“We see (Grafana Machine Learning) as a very useful tool for intelligent anomaly detection, and it will certainly become one of the tools that our SREs will use to increase their productivity and reduce their daily toil.”

Forecasting

Wed, 17 Jun 2026 12:08:02 -0500

Forecasting

Forecasting creates smarter alerting. It learns a metric’s past pattern and projects its future values. When the forecast shows the metric is on track to cross a limit, an alert fires before you actually hit it.

The outcome is:

Early detection of issues before they become real problems
Capacity planning based on data instead of guesswork

Try it out in Grafana Play

Outlier detection

Wed, 17 Jun 2026 12:08:02 -0500

Outlier detection

If you’re watching average CPU across a 20-node cluster:

One node at 95% barely moves the cluster average, so nothing looks wrong.
An alert doesn’t fire, so you don’t know until it affects users.

When you’re trying to determine an instance that is behaving differently from its peers, a fixed threshold can’t help.

Outlier detection solves these kinds of issues.

You define the group of instances that should behave alike.
Outlier detection flags any that drift from the rest.

Try it out in Grafana Play