Documentation Grafana Cloud AI and machine learning Grafana Assistant Get started

Get started with Grafana Assistant

This article shows you how to set up Grafana Assistant in Grafana Cloud or in self-hosted Grafana, get started using its core workflows, and understand which features are available in each deployment model.

Grafana Assistant is an AI-powered observability agent for Grafana that helps you monitor, troubleshoot, and manage your systems through natural language conversations. Instead of learning complex query languages or memorizing dashboard locations, you can ask simple questions like “Show me CPU usage” or “Create a dashboard for my database.”

The Assistant understands your observability data across metrics, logs, traces, profiles, and databases. This makes it easier to gain insights and solve problems without requiring deep technical expertise in each tool.

What Grafana Assistant can do

Grafana Assistant acts as your observability expert, helping you work more effectively with your monitoring data through five core workflows:

• Run investigations: Ask questions about performance, launch investigations, and correlate metrics, logs, traces, profiles, and SQL data.
• Manage dashboards: Create new monitoring dashboards or refine existing panels, layouts, and variables.
• Query data: Build and refine PromQL, LogQL, TraceQL, SQL, and k6 queries with validation and optimization tips.
• Navigate Grafana resources: Find dashboards, data sources, and related tools without leaving the conversation.
• Knowledge and best practices: Ask for Grafana guidance, observability concepts, and monitoring strategies in context.

Choose your deployment model

Assistant supports two setup models:

• Grafana Cloud: Assistant runs directly in your Grafana Cloud stack.
• Self-hosted Grafana (OSS or Enterprise): Install the Assistant app in your own Grafana instance and connect it to a Grafana Cloud stack. Grafana Assistant on-premise has a reduced feature set compared with the full Grafana Cloud experience.

Before you begin

Make sure you have the right stack, access, and data before you start.

• Grafana Cloud stack: You need a Grafana Cloud stack with Assistant enabled. Self-hosted Grafana deployments also require a Grafana Cloud stack because the Assistant backend, limits, and billing are handled there.
• Administrator access: You need administrator permissions in the Grafana deployment where you enable or connect Assistant. The self-hosted connection flow also requires an organization admin in the self-hosted Grafana instance.
• Data sources (recommended): While not required to get started, the Assistant works best when you have monitoring data configured in your Grafana deployment. Supported data sources include:
  • Prometheus: Collects and stores metrics such as response times, CPU usage, memory consumption, and request rates from your applications and infrastructure.
  • Loki: Aggregates and searches logs such as application error messages, access logs, system events, and debugging information.
  • Tempo: Traces requests across distributed systems to identify bottlenecks, latency issues, and service dependencies in microservices architectures.
  • Pyroscope: Continuously profiles application performance to identify memory leaks, CPU hotspots, and inefficient code paths.
  • k6: Provides performance testing results including load test metrics, user simulation data, and API response benchmarks.
  • SQL databases: Supplies business and operational data from PostgreSQL, MySQL, ClickHouse, BigQuery, and more for correlating metrics with business outcomes.
  • On-premises data sources available through Grafana Cloud connections.

Enable Grafana Assistant in Grafana Cloud

Enable Assistant in the Grafana Cloud stack that will back your deployment.

An administrator must accept the terms and conditions to enable Assistant for your stack:

1. Sign in to Grafana Cloud as an administrator.
2. Navigate to Administration > Plugins and data > Plugins.
3. Search for Grafana Assistant or go directly to /plugins/grafana-assistant-app.
4. Review the terms and conditions.
5. Select the agreement checkbox.
6. Click Save.

After you enable the Assistant, the sparkle icon appears in the top-right corner of Grafana for all users in your stack.

Connect a self-hosted Grafana instance

If you run Grafana OSS or Grafana Enterprise, install the Assistant app in that Grafana instance and pair it with a Grafana Cloud stack.

1. Enable Grafana Assistant in the Grafana Cloud stack that will back the deployment.
2. Install the Grafana Assistant app in your self-hosted Grafana instance. In Grafana, search for Grafana Assistant. The plugin ID is grafana-assistant-app.
3. Sign in to the self-hosted Grafana instance as an organization administrator.
4. Open Administration > Plugins and data > Plugins and select Grafana Assistant.
5. Open Connection.
6. Select Start connection and complete the authorization flow in Grafana Cloud.

Make sure Grafana can load the grafana-assistant-app plugin before you start the connection flow.

If you prefer to configure the connection manually, enter the following values from your Grafana Cloud stack in Connection:

• Backend URL
• Instance ID
• Access Token

After the connection succeeds, the page reloads and Assistant becomes available in the self-hosted Grafana UI.

Understand on-premise feature differences

Grafana Assistant on-premise keeps the main chat workflows available.

You can use on-premise Assistant for chat, dashboarding, querying, navigation, rules, quickstarts, and skills.

For the full feature list, including features that aren’t available on-premise, refer to Use Grafana Assistant on-premise.

Limit Assistant to specific stacks

Assistant is available in all Grafana Cloud stacks by default. Within a stack, use RBAC to control who can open Assistant by scoping plugins.app:access to plugins:id:grafana-assistant-app.

Opt out or disable after testing

If you tried Assistant and prefer not to keep it enabled, you have several options:

• Do nothing if you never enabled it: When not enabled by an administrator, Assistant features remain off and you don’t send data to AI providers.
• Restrict access with RBAC: Limit who can open Assistant in a stack by scoping plugins.app:access to plugins:id:grafana-assistant-app for only the roles you choose. Refer to Manage Assistant access with RBAC.
• Clean up local usage: Users can delete conversations in the Assistant UI. Administrators can disconnect MCP servers and OAuth integrations that were added for testing.
• Disable completely: To do this, an administrator can navigate to Administration > Plugins and data > Plugins, search for Grafana Assistant or go directly to /plugins/grafana-assistant-app, clear the agreement checkbox, and click Save.

Try the tutorial

Grafana Assistant includes an interactive tutorial that guides you through the main features and workflows. The tutorial helps you get hands-on experience with the most important Assistant workflows.

When you first open Grafana Assistant, you’ll see the tutorial option. The tutorial takes you on a journey through:

• Asking questions about your observability data
• Creating and modifying dashboards
• Building queries across different data sources
• Using mentions to get more specific results
• Navigating your Grafana resources

The tutorial is interactive. You’ll practice each feature directly within the Assistant interface. You can complete the entire tutorial in one session or return to it later. To access the tutorial at any time, click Menu (three vertical dots) in the Assistant chat and select Open tutorial.

Start your first conversation

Open Assistant in the top-right corner of Grafana or select Assistant from the left navigation menu. This opens a chat interface where you can ask questions in natural language.

The Assistant works best when you’re specific about what you want to achieve. Think of it as talking to a monitoring expert who has access to all your data but needs context about what you’re trying to accomplish.

Your first query

Start with a simple question to explore your data. For example:

Copy

Show me CPU usage for the API service

When you ask this question, several things happen automatically:

1. Data source detection: The Assistant identifies which of your configured data sources contain CPU metrics.
2. Query generation: It creates the appropriate queries, for example, PromQL for Prometheus data.
3. Multi-signal analysis: It correlates CPU metrics with related logs, traces, and other data to provide context.
4. Visualization: It presents the results in charts, graphs, or tables as appropriate.
5. Insights: It highlights patterns, anomalies, or trends in the data.

This process demonstrates the Assistant’s core strength: turning complex observability tasks into simple conversations.

Try different types of requests

Once you’re comfortable with basic queries, explore the Assistant’s other workflows:

• Dashboard creation: “Create a dashboard for monitoring my database performance.”
• Query building: “Help me build a query to find all error logs from the last hour.”
• Navigation: “Show me dashboards related to Kubernetes monitoring.”
• Troubleshooting: “Why is my application responding slowly?”
• Knowledge: “How do I set up alerting for high memory usage?”

Each request type uses the Assistant’s different workflows, from analyzing your specific data to providing general observability guidance.

Use mentions for better results

The @ mention feature helps the Assistant provide more targeted and accurate responses by adding specific context to your questions. This feature is particularly useful when you have multiple data sources or when you want to focus on specific resources.

To use mentions, type @ in your message or click Mention (@) in the chat interface. This opens a menu where you can select from:

• Data sources: Specific monitoring tools like your Prometheus instance, Loki logs, or Tempo traces.
• Dashboards and panels: Existing visualizations you want to reference or modify.
• Other Grafana resources: Folders, alerts, or other configured items.

Example: Generic vs. specific questions

Instead of asking a generic question like:

Copy

Show me error rates

Use mentions to be specific:

Copy

Show me error rates from @my-application-logs

The specific version helps the Assistant understand exactly which data source to query, what format to expect, and what context to provide in the response. This results in faster, more accurate answers tailored to your specific setup.

Provide feedback to improve the Assistant

Assistant improves through user feedback. Refer to the privacy and security documentation to learn how Assistant works with your data. After each response, you see thumbs up and thumbs down icons that help improve answer quality.

When to use thumbs up:

• The response accurately answered your question
• The generated queries or dashboards work as expected
• The explanation was clear and helpful
• The Assistant found the right resources or data

When to use thumbs down:

• The response was inaccurate or incomplete
• Generated queries had errors or didn’t work
• The Assistant misunderstood your question
• The information provided wasn’t relevant to your needs

When you give a thumbs down, you can often provide additional context about what went wrong. This specific feedback is especially valuable for improving the Assistant’s performance.

Your feedback helps make Assistant more effective for you and other users.

Next steps

Now that you have Grafana Assistant enabled and understand the basics, here are some ways to get the most value from your observability setup:

Immediate actions:

• Start with simple queries about your most important systems or applications
• Ask for a tour of your existing dashboards: “What dashboards do we have for database monitoring?”
• Identify gaps: “What metrics should we be monitoring for our web application?”

Build your skills gradually:

• Break down complex investigations into smaller questions, for example, “Show me error rates” followed by “What’s causing these errors?”
• Experiment with different question formats to see how the Assistant interprets your needs.
• Use the Assistant as a learning tool by asking “How do I…” questions about observability concepts.

Explore workflows:

• Run investigations: Ask questions about performance, launch investigations, and correlate observability signals.
• Manage dashboards: Create new monitoring dashboards or refine existing panels, layouts, and variables.
• Query data: Build and refine PromQL, LogQL, TraceQL, SQL, and k6 queries with validation and optimization tips.
• Navigate Grafana resources: Find dashboards, data sources, and related tools without leaving the conversation.
• Knowledge and best practices: Ask for Grafana guidance, observability concepts, and monitoring strategies in context.

Remember: Grafana Assistant grows with your needs. Start simple, learn what works for your environment, and gradually explore more sophisticated monitoring and troubleshooting workflows. If you use self-hosted Grafana, keep the on-premise feature boundary in mind when following Grafana Cloud-focused guides.