Skip to content
PERCONA MONITORING AND MANAGEMENT DOCUMENTATION

Rate this page
☆ ☆ ☆ ☆ ☆
Thanks for your feedback
Thank you! The feedback has been submitted.

Get free database assistance or contact our experts for personalized support.

Get help from Percona

AI features in PMM (ADRE, Investigations, QAN AI Insights)

PMM can connect to HolmesGPT so you can use AI-assisted analysis alongside metrics, Query Analytics (QAN), and dashboards. This section is written for DBAs and SREs who use the PMM UI.

For operators (URLs, APIs, Holmes configuration, Grafana panel rendering): see the developer docs linked at the end of this page.

When to use which feature

Feature Use it when you want…
ADRE Chat General chat — quick questions, context-aware help, and conversation about your environment while you work in PMM.
Investigations A deep dive with a structured report — blocks, timeline, running a full investigation, exporting PDF, and optionally creating a ServiceNow ticket if your admin configured it.
QAN AI Insights Query optimisation and tuning — analysis focused on a specific query pattern in QAN, distinct from open-ended chat.

Glossary

Term Meaning
ADRE Autonomous Database Reliability Engineer — PMM’s name for the AI assistant integration (including ADRE Chat in the UI).
HolmesGPT The analysis backend PMM calls for many AI operations. Your organisation runs HolmesGPT where it can reach PMM APIs and (if configured) other data sources.
Fast vs Investigation In ADRE Chat, Fast favours quick answers with lighter Holmes skills / TodoWrite usage by default; Investigation uses the full investigation-oriented controls. Admins tune Holmes behavior_controls and prompts under AI Assistant settings.
Investigation A persisted incident page: title, status, blocks (findings, markdown, panels, query results, etc.), comments, and messages.
Block A typed piece of content inside an investigation report (for example summary, finding, or slow-query analysis).
QAN AI Insights AI-generated optimisation guidance for QAN data, with server-side caching per query and service.

Privacy and networking

  • Messages and investigation runs are processed by the configured backend (typically HolmesGPT). Treat prompts and responses according to your organisation’s data policy.
  • Use TLS and network policies so traffic between PMM and HolmesGPT is protected. Do not embed real passwords or API keys in URLs; store secrets in PMM settings or your secret manager as your admin defines.
  • PMM may send Grafana URL context (path, variables, optional tab title) with ADRE Chat so the assistant knows which dashboard view you are on. That context is descriptive metadata, not a substitute for access control.

Where to configure

  • Configuration → Settings (Advanced / AI-related options) and the AI Assistant / ADRE area in the UI (exact labels depend on your PMM build).
  • ServiceNow (optional): URL, API key, and client token are set in the same settings area when your organisation enables ticketing. Never share these values in chat or documentation.

Further reading (technical)

Source files live in the PMM repository (not all are part of the published MkDocs tree):