Documentation Grafana Cloud AI and machine learning Grafana Assistant (Preview) Features Infrastructure Memory

Infrastructure Memory

Infrastructure Memory enables Grafana Assistant to build a comprehensive understanding of your infrastructure by discovering services, analyzing their structure, and storing detailed reports about how they operate and are monitored.

This pre-analyzed knowledge helps the Assistant and Investigations answer your questions faster and with better precision by providing essential context about your environment.

To run Infrastructure Memory, navigate to Settings and select the Infrastructure Memory tab, where you can manually initiate the discovery and analysis process.

How it works

Infrastructure Memory operates in two phases to build a complete picture of your infrastructure. Memories are created per datasource, and currently only Prometheus datasources are supported.

Phase 1: Service discovery

The discovery process starts by querying your configured Prometheus datasources to find all services running in your environment.

During discovery, an agent:

• Runs queries across your Prometheus datasources
• Identifies services and components using labels and metadata
• Groups discovered entities by functionality or business domain (for example, “Mimir”, “Shopping Cart”, “User Authentication”)
• Stores the discovery results for analysis

This phase creates an inventory of what runs in your infrastructure and how it’s organized.

Phase 2: Group analysis

After discovery, specialized agents analyze each service group in parallel to create detailed infrastructure memories.

A swarm of agents works concurrently to:

• Analyze deployment patterns and infrastructure layouts
• Document where services run (clusters, namespaces, nodes, regions)
• Identify how services are monitored (Prometheus, OTLP, Alloy, tracing systems)
• Discover dependencies between services and components
• Capture resource characteristics and operational baselines
• Document important metrics, labels, and instrumentation patterns

Each agent produces a comprehensive report containing:

• Functionalities - What each service group does and its purpose
• Structure - How services are deployed and organized
• Dependencies - Internal and external service relationships
• Monitoring - Available metrics, traces, logs, and how to query them
• Important metrics and labels - Key observability signals and useful label patterns

These reports are stored as infrastructure memories.

Benefits

Infrastructure Memory provides several advantages for observability workflows:

• Faster responses - The Assistant has immediate access to infrastructure context without needing to discover it during each conversation
• Better precision - Queries use actual service names, namespaces, and metric names from your environment
• Reduced exploration time - Agents know where to look and what tools to use based on stored knowledge
• Improved investigations - Investigations start with comprehensive context about relevant services and their monitoring

How the Assistant uses infrastructure memory

When you interact with Grafana Assistant or start an Investigation, infrastructure memories provide context in two ways:

Retrieval by name

If you mention specific service groups, the Assistant retrieves those exact memories:

Copy

Tell me about the shopping cart service

The Assistant retrieves infrastructure memory for the shopping cart group and uses it to answer your question with precise details about that service.

Retrieval by intent

For broader questions, the Assistant analyzes all groups and extracts relevant information:

Copy

What services handle user authentication?

The Assistant examines all infrastructure memories to identify and extract information about authentication-related services, even if you don’t know the exact service names.

Infrastructure memory in investigations

Investigations use infrastructure memories to:

• Provide specialists with precise service names, namespaces, and datasource UIDs
• Understand service dependencies and potential blast radius
• Design targeted parallel tasks based on actual service groupings
• Avoid assumptions about label names, metric names, and configurations

This context helps investigation agents work more efficiently and produce more accurate root cause analysis.

What’s next?

Infrastructure Memory works seamlessly with other Grafana Assistant features to enhance your observability experience:

• Learn how to use Assistant chat with infrastructure context
• Discover how Investigations leverage infrastructure memories
• Read the Get started guide to begin using Grafana Assistant