Skip to main content

AI Chat with Production

Overview​

AI Chat is an interactive interface in the OpsWorker portal that lets you ask questions about your Kubernetes clusters and get answers grounded in live cluster data. Unlike generic AI chatbots, OpsWorker Chat has direct access to your infrastructure through the installed Kubernetes Agent and any configured integrations.

How It Works​

  1. Open Chat in the OpsWorker portal
  2. Select the cluster you want to query
  3. Ask a question in natural language
  4. OpsWorker's AI agents query your cluster and integrations in real time
  5. You receive an answer based on actual, current data

Chat uses a WebSocket connection for real-time, streaming responses.

Capabilities​

Two Chat Modes​

ModeContextUse When
Investigation ChatScoped to one completed investigationYou want to dig deeper into a finished investigation
Free-form Chat ("AI SRE")General cluster, metrics, and code queriesYou want to explore cluster state, check health, or debug across data sources

Investigation Chat answers from the stored state of a single completed investigation (the resources, logs, and events already gathered). It has an optional per-session Live Queries toggle that lets it run fresh real-time reads against the cluster when stored state is not enough.

Free-form Chat is multi-agent. A tool router (Amazon Nova) prefilters the available tools for your question, then an orchestrator agent delegates to sub-agents through an invoke_agent tool. Which sub-agents are available is gated by the integration capabilities connected for the org and cluster (see Multi-Agent Workflows).

What You Can Ask​

  • Cluster health: "Are there any pods in CrashLoopBackOff?"
  • Resource inspection: "Show me the deployment config for api-gateway"
  • Log analysis: "What errors are in the logs for payment-service?"
  • Topology: "What services depend on the database?"
  • Change tracking: "What changed in namespace production recently?" (with Git integration)
  • Metrics: "What's the CPU usage for this namespace?" (with Grafana MCP integration)
  • Right-sizing: "Is the worker deployment over-provisioned?" (with Grafana MCP metrics + cluster access)
  • Diagrams: "Draw me a topology diagram of the payment flow"

Capability-Gated Agents​

Free-form Chat exposes sub-agents based on the integration capabilities connected for the cluster and org. The capability tokens are cluster, sourcecontrol, and metrics:

  • Cluster reads (cluster): Available once the in-cluster Kubernetes Agent is installed; queries pods, services, logs, events, and configs, and validates and maps resource dependencies.
  • Source code (sourcecontrol): Available with a GitHub or GitLab integration; reads commits, PRs/MRs, and file contents.
  • Metrics (metrics): Available with the Grafana MCP integration; runs Prometheus (PromQL) queries through a Grafana datasource.
  • Resource optimizer: Available when both cluster and metrics are connected; produces right-sizing and cost recommendations.

Sub-agents whose capabilities are not connected are shown as unavailable with their missing_capabilities listed. See Multi-Agent Workflows for the full roster.

Read-Only with One Write Path​

Chat is read-only against the cluster. It never executes kubectl mutations or applies changes. The only write path is that the source code and resource optimizer agents can open a Git pull request / merge request with a proposed change, which a human reviews and merges. OpsWorker never merges on its own.

Diagrams and Charts​

  • Mermaid diagrams: Chat auto-generates topology and flow diagrams (for example, "draw me a diagram of the payment flow").
  • Time-series charts: Prometheus query results are auto-charted inline.

Memory in Chat​

During a conversation the AI can store and recall AI Memory. It uses tools such as remember_preference, remember_cluster_fact, promote_personal_note, and forget_memory, and it auto-extracts facts, preferences, and conversation summaries so context persists across sessions.

Conversation Features​

  • Multi-turn: Maintain context across multiple questions
  • Prompt suggestions: Pre-built prompts to help you get started
  • Agent execution blocks: See which agents were invoked and what data they retrieved
  • Conversation history: Previous conversations are saved and accessible from the sidebar

Next Steps​