Grafana Labs
Also known as: Grafana Assistant, Grafana Investigations, Grafana Cloud
Grafana Assistant helps site reliability and operations teams investigate incidents using data already connected to Grafana. An investigation can start from a prompt, a monitoring alert, or an incident webhook. The agent queries telemetry, compares possible causes, and writes findings with the queries and charts that support them. Teams can redirect a running investigation, connect external tools, and save recurring checks that alert engineers when the behavior of a service changes.
Grafana Assistant is the observability agent in Grafana Cloud. It works with telemetry such as Prometheus metrics, Loki logs, and Tempo traces. When an investigation starts, the agent gathers evidence, tests possible causes, and writes a report. A responder can inspect which queries support each finding and open the panel behind the result. The report separates findings, hypotheses, and sources. Engineers can correct an assumption and redirect a running investigation if they learn that it started with the wrong service or time window.
Teams can start investigations by hand or configure them to start from alerts or incident webhooks. Watchers run recurring checks and learn a baseline for the service or workload they monitor. A Watcher can launch an investigation when conditions warrant it. Teams can also save automations that run a defined prompt on a schedule or in response to an event. The Assistant uses a knowledge graph to retain information about services and their dependencies. Operators can provide team runbooks as skills and connect remote MCP tools to look up issues or propose changes in an external system. Grafana Pages and Managed Sandboxes extend the Assistant in public preview.
Grafana Cloud Free includes up to three active Assistant users, with 40 million tokens per user per month and a separate 25 million token pool for system work. The Pro platform starts at $19 per month with the same allowance included. Additional active users start at $20 each, and additional tokens start at $2 per million. Investigations use tokens and carry no separate fee.
Enterprise runs on an annual commitment with a $25,000 minimum. A self managed Grafana instance can connect to the Grafana Cloud backend for chat and dashboards. The backend still processes prompts and query context, and some workflows, including Investigations and Watchers, require Grafana Cloud. Grafana Cloud carries SOC 2 Type II, PCI DSS and FedRAMP compliance.
Grafana Labs sells the Assistant into a large observability base: Gartner named it a Leader in its 2026 Magic Quadrant for Observability Platforms, and Microsoft, Salesforce, Uber, NVIDIA, AXA, BlackRock, Delivery Hero and LATAM Airlines run Grafana. LexisNexis Risk Solutions uses the Assistant to reach root cause faster without deep expertise in every part of a complex system, and MasterControl and Transmute Data are Assistant customers too. An operations team uses Grafana Assistant most directly when the same stack already brings together the telemetry that on call engineers would need to inspect.
A team can assign access to investigations and data sources through Grafana roles and share a report with colleagues who handle the affected service. An engineer can follow a finding into its original query, add evidence from a deployment, and ask for the report to reflect that evidence. A report can be exported for an incident review or handed to a service owner with the relevant panels attached. Reports in Workspace can be downloaded as Markdown, HTML, or PDF.
Vendor details
Canonical URL
https://grafana.com
Category
SRE / DevOps agent
Subcategory
Incident investigation and agent led observability
Funding status
Grafana Labs Inc. is a privately held software company.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Grafana Assistant works with Grafana data sources, alerting, and incident response. Administrators can connect remote MCP servers and bring Grafana tools into external agents.
In practice
An alert for checkout latency starts an investigation that compares recent traces and database metrics and presents the supporting queries to the on call engineer.
A platform team schedules a Watcher to check the telemetry of a payment service and investigate changes beyond its calibrated baseline.
A responder adds an issue tracker through MCP to look up related incidents, review the agent's suggested ticket changes, and approve the action.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
11.5 / 14 capabilities · 82%
| Integrations & Tool Calling | Full |
|---|---|
|
Configured remote tools let the Assistant use operational information and take actions in external systems, such as opening or updating tickets. Administrators enable integrations and users can connect tools with their own OAuth identity. Reusable skills can tell the agent when a particular external tool is relevant and supported calls present confirmation controls. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/configure/mcp-serversread 2026-11-20 |
|
| Workflow Orchestration | Full |
|
An investigation coordinates multiple steps against the connected telemetry. The agent develops possible causes, requests additional information, queries sources, and writes a report with evidence. A saved automation or a Watcher can initiate a new run. Operators can continue an existing investigation with new evidence and the workflow updates the hypotheses and findings. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/platform/investigationread 2026-11-20 |
|
| Knowledge Grounding & RAG | Full |
|
Investigations query the buyer's connected data sources and cite the evidence for a possible root cause. A team can inspect a citation, open the underlying query, and compare the selected range with the other evidence. The product also uses a knowledge graph of infrastructure and can retrieve documentation or issue context through configured remote tools. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/platform/investigationread 2026-11-20 |
|
| Human Oversight & Guardrails | Full |
|
External actions in published Pages can prompt an operator to review destination and values before approving. Remote MCP usage also supports manual confirmations. During investigations, operators can revise hypotheses, redirect the agent when an assumption changes, or stop a run. Permissions and tool settings govern the action surface a given team can use. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/configure/pagesread 2026-11-20 |
|
| Security, Identity & Governance | Full |
|
Assistant enforces Grafana data source permissions and inherits role based access control. Administrators can delegate privileges for investigations, conversation access, and managing tool servers. Grafana Cloud carries SOC 2 Type II, PCI DSS and FedRAMP compliance, and conversation content is not used to train outside providers' models. The same controls apply to a connected self managed interface. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/privacy-and-security/securityread 2026-11-20 |
|
| Observability & Auditability | Full |
|
A responder can review hypotheses, generated panels, queries, sources, and a timeline for an agent run. Reports can be exported and existing investigations can be resumed with new evidence. Usage is broken out by user, service account and stack in billing. Operators can see how reported conclusions were assembled and inspect the source behind one finding. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/platform/investigationread 2026-11-20 |
|
| Memory & State Persistence | Full |
|
The Assistant automatically discovers connected sources and refreshes a persistent record of services and dependencies. That record can inform later requests across conversations and can be searched. Source permissions limit which records a user can see. Investigation history and follow up runs also retain the evidence and conclusions from a previous run. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/platform/infrastructure-memoryread 2026-11-20 |
|
| Deployment & Data Residency | Partial |
|
Organizations can deploy the Grafana interface themselves and keep underlying data sources on their own infrastructure. A self managed Assistant still connects to a cloud stack that processes the supplied prompt and query context. Some agent features only work inside Grafana Cloud, including full investigations, scheduled Watchers, and hosted MCP connections. Sourcegrafana.com/docs/grafana/latest/ai-tools/grafana-assistant/self-managedread 2026-11-20 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
Skills ship with dedicated assistant modes, including Incident Investigation and k6 Script Authoring. Operators can add their own resources and define reusable prompts in team quickstarts. Watchers are managed agent workflows a team configures from a plain description and points at existing telemetry. Sourcegrafana.com/grafana/plugins/grafana-assistant-app/changelogread 2026-11-20 |
|
| Triggers & Channel Coverage | Full |
|
The Assistant can start from a manual request, Grafana alerts, incident webhooks, or an automatically scheduled Watcher. Automations can run on schedule or in response to an external event. Operators can review conversations in Grafana Workspace and supported collaboration clients, where personal and system triggered accounts are kept apart. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/pricingread 2026-11-20 |
|
| Model Flexibility & Routing | Partial |
|
Operators can leave selection on Auto or pick the capability tier used for a later turn in an investigation. Comparable choices are available in some Workspace chats. The available tier controls make it possible to change the resource setting, while the included model providers and routing across them are managed by Grafana. Records produced by the run remain associated with the investigation and its telemetry. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/platform/investigationread 2026-11-20 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
External agents can connect to a hosted MCP server to look up Grafana resources and work with supported incident and investigation functions. An administrator can scope the tool areas exposed on that connection. An HTTP API serves assistant quickstarts, and Terraform can manage Assistant resources as code. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/configureread 2026-11-20 |
|
| Testing, Debugging & Optimization | Partial |
|
Watchers expose calibration before a team relies on recurring checks. An engineer can revise the selected inputs and thresholds, inspect investigation evidence, and record whether a causal hypothesis held up. There is no team owned evaluation suite with a fixed dataset and expected outputs to run against the agent. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/platform/watchersread 2026-11-20 |
|
| Browser / Computer-use | Not documented |
|
The Assistant can help produce k6 scripts and hand the result off to the testing surface. Separate k6 synthetic checks can control a headless browser under the test script. The investigation agent itself works through queries and attached tools, with no browser or desktop session of its own for general actions. Sourcegrafana.com/docs/grafana-cloud/platform/grafana-assistant/guides/k6-script-authoringread 2026-11-20 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Free for up to 3 active AI users, Pro platform from $19 a month
Base platform fee, active AI users, additional tokens
What is public
Grafana Cloud Pro starts at $19 a month and includes three active AI users, 40 million monthly tokens for each user, and 25 million monthly tokens for work started by the system. The Free plan limits the number of users and tokens.
Billing mechanics
Additional active users start at $20 and additional tokens at $2 per million. An automated investigation uses tokens from a shared pool. A manual investigation uses the initiating user's allowance.
Variable cost rationale
Past the three included users and their token allowances, every added active user and every million tokens bills on top of the $19 platform fee, so automated investigations and Watchers can raise spend without anyone opening a chat.
Additional watchouts
Connecting to hosted MCP counts as active usage even if no tokens are consumed. Watchers draw from included system tokens starting December 1, 2026.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free plan with up to 3 active AI users and token limits
Lowest paid plan
Grafana Cloud Pro
Missing data
Enterprise has a $25,000 annual minimum, and its unit prices are set against the annual commitment.
Related vendors
- AlertD — AI agents for AWS operations that run read only in the customer's…
- Anyshift — AI SRE built on a versioned knowledge graph of infrastructure, apps…
- Avesha — Obliq, Avesha's autonomous AI SRE for Kubernetes and agentic…
- AWS DevOps Agent — AWS DevOps Agent investigates incidents and proposes mitigations for…
- Better Stack — Better Stack offers incident management with built in on call,…
- BigPanda — BigPanda is an agentic IT operations platform whose Biggy agents…
Alternatives to Grafana Labs
Other products that do the work Grafana Labs does, and what each one is.
- Edge DeltaAgentic observability platform
- BigPandaAgentic IT operations platform
- RootlyAI native incident management platform
- AWS DevOps Agent
- DynatraceObservability platform
- incident.ioIncident management platform in Slack and Teams