ADRE Chat¶
ADRE Chat is the floating chat in the PMM UI for general conversation with the AI assistant: questions about your environment, metrics, alerts, and how to interpret what you see.
It is different from Investigations (structured reports and deep runs) and QAN AI Insights (query-focused tuning).
Requirements¶
- An administrator must enable ADRE and set the HolmesGPT base URL in PMM AI Assistant settings.
- HolmesGPT must be reachable from the PMM server over your network. Use HTTPS in production where possible.
What gets sent to the backend¶
- Your messages and a short window of recent chat history.
- A system preamble that includes PMM context. When you are on Grafana routes, PMM may attach current Grafana context derived from the URL synced with the Grafana iframe (dashboard UID, optional focused panel, time range, template variables, and sometimes the browser tab title). That helps the assistant answer “what am I looking at?” without guessing.
Fast vs Investigation¶
The ADRE panel can run in Fast or Investigation mode. PMM sends Holmes behavior_controls and an additional_system_prompt appropriate to the mode (tunable under Configuration → AI Assistant). Operators should be aware that the Holmes container environment variable ENABLED_PROMPTS can override what the API may enable.
Technical details: dev/adre/README.md and Holmes fast mode / prompt controls.
Good practices¶
- Do not paste passwords, connection strings with secrets, or API keys into chat.
- If answers reference panel images, PMM may show rendered graphs; technical details of the image proxy are documented under developer docs only.