Understand infrastructure memories

Infrastructure memories capture snapshots of your services, dependencies, and monitoring signals so Grafana Assistant answers with accurate context. This article explains what memories contain, how discovery scans build them, and how they support every workflow that relies on topology knowledge.

Before you begin

Grafana Cloud organization: Enable Grafana Assistant and investigations.
Prometheus data sources: Discovery currently scans Prometheus metrics to infer services and dependencies.
Permissions: You need access to run discovery scans from the Assistant settings page.

What infrastructure memories contain

Each memory groups related services and stores an overview, infrastructure layout, monitoring signals, label selectors, and a triage checklist. That combination gives the Assistant enough information to describe the service, highlight its dependencies, suggest relevant metrics or logs, and remind you which labels matter during incidents. The discovery workflow saves the memories in Grafana Cloud so the Assistant can fetch them whenever you mention a service or request intent-based context.

How discovery scans work

From Grafana Assistant > Settings > Infrastructure memory, start a scan and choose Prometheus data sources.
Discovery queries those data sources for service metadata, aggregates related labels, and builds memories.
Memories replace any older data for the selected data sources. Plan scans around deployments or major infrastructure changes.

Behind the scenes, the Assistant conversation manager loads memories asynchronously and injects them into the chat history the next time you mention a service.

Keep memories accurate

Run routine scans or trigger them after major changes so new services and labels appear immediately. Review the generated memories, note gaps in your incident runbooks, and remove entries that describe deprecated systems. When a scan fails, retry with a smaller data source set to isolate the issue before rerunning the full job.

How memories support workflows

In chat, mentioning a service unlocks its memory so responses include accurate dependencies and monitoring hints.

Investigations use the same data when the investigation agents plan their tasks, which keeps them aligned with the actual topology.

Navigation prompts benefit as well because discovery-based context helps the Assistant list the dashboards and data sources most relevant to the incident at hand. When you ask for new panels or variables, the memory provides the correct labels so dashboard management stays consistent.