The module lifecycle stage: General Availability
The module has requirements for installation

Overview

The AI assistant is a chat helper built into the DOP web interface and powered by a large language model (LLM). The assistant works with your project data and can:

  • answer questions about the project’s metrics, logs, traces, and alerts (“what happened to CPU on the servers over the last day?”, “why did this trigger fire?”);
  • write and explain PromQL, LogQL, and TraceQL queries;
  • analyze active alerts and notification history, and help investigate incidents;
  • create and modify platform objects — alerting rules, dashboards, silences — previewing and confirming every change first;
  • answer “how do I do … in DOP” questions using the built-in documentation, including the agent metrics reference;
  • help configure opAgent by suggesting ready-to-use commands and configuration files.

The assistant acts strictly on behalf of the current user: it only sees the workspaces, projects, and data the user has access to, and cannot change anything the user has no permission to change.

Note: user messages and the platform data the assistant reads to answer (metrics, logs, configuration fragments) are sent to the external LLM at the address configured during setup. Keep this in mind when choosing an LLM provider.

Enabling

The assistant is disabled by default. To enable it, create the llm-assistant Secret in the platform namespace:

apiVersion: v1
kind: Secret
metadata:
  name: llm-assistant
  namespace: d8-observability-platform
stringData:
  LLM_BASE_URL: "https://llm.example.com/v1"
  LLM_MODEL: "gpt-4o"
  LLM_API_KEY: "sk-..."

Secret keys:

Key Required Description
LLM_BASE_URL yes OpenAI-compatible API address (with the /v1 suffix)
LLM_MODEL yes Model name
LLM_API_KEY yes API key
LLM_MAX_TOKENS no Maximum tokens in a model response (default 40000)
LLM_CONTEXT_WINDOW no Model context window size; 0 — unlimited
LLM_MAX_CONTEXT_TOKENS no Context budget per conversation (default 300000)
LLM_CUSTOM_CA no Custom CA PEM bundle if the LLM endpoint uses a private certificate
DISABLE_WRITE_OPS no "true" — forbid the assistant from making any changes installation-wide

Once the secret is created, the platform automatically deploys the assistant service and enables the button in the interface (the rollout takes a few minutes). The assistant also requires an external PostgreSQL database for conversation history — no extra steps are needed on installations that already have the external database configured.

Changes to the secret keys are applied automatically by restarting the service. To disable the assistant, delete the secret.

Usage

  • The assistant opens via the AI button in the top bar, the floating button in the bottom-right corner, or the Ctrl/Cmd+I shortcut.
  • The chat window supports multiple conversation tabs and a history: a closed tab stays in the history, where it can be reopened or deleted permanently.
  • While working, the assistant shows its reasoning and the steps it takes (which metrics it queries, which documentation pages it reads) in the expandable “Thinking…” block.
  • The assistant is aware of the page context: with a project open, questions refer to it by default.

Changes and audit

By default the assistant is read-only. To let it create and modify objects, enable the “Allow changes” toggle in the chat window. Even with the toggle on, the assistant first shows exactly what it is about to change and applies the change only after an explicit confirmation in the conversation.

Every applied change is recorded:

  • a “Change applied” plate with a link to the created or modified object appears in the conversation feed;
  • operations are marked as performed via the assistant in the platform audit log;
  • a structured line with the user, project, and action is written to the server log.

The operator can forbid changes via the assistant installation-wide with the DISABLE_WRITE_OPS: "true" key in the llm-assistant secret — the interface toggle then has no effect.