Skip to main content

AI Agents

What are AI Agents

OpsWorker uses specialized AI agents to investigate alerts and answer questions about your infrastructure. Each agent has access to specific tools and data sources, and is gated by the integration capabilities you have connected for the cluster.

During investigations and chat sessions, OpsWorker automatically selects which agents to invoke based on the question being asked and which capabilities are available.

Capabilities

Agents are unlocked by three integration capabilities. If a capability is missing, the agents that depend on it are shown as unavailable.

CapabilityUnlocked byPowers
clusterKubernetes Agent installedKubernetes data access (pods, logs, events, deployments, services, ingresses)
sourcecontrolGitHub or GitLab integrationSource code and repository analysis, branch and pull/merge request creation
metricsGrafana MCP integrationMetrics and dashboard queries (Prometheus via the Grafana datasource)

Free-form AI Chat agents

In free-form AI Chat, an orchestrator routes your question to one or more specialized sub-agents:

AgentRequiresWhat it does
Investigate IssueclusterFocused symptom investigation from logs, events, and resource state
Analyze LogsclusterPod log error and anomaly analysis
Validate ResourcesclusterReviews CPU/memory requests and limits, replicas, and health checks
Check DependenciesclusterMaps service dependencies via services, endpoints, and network policies
Source Code & RepositorysourcecontrolCross-repo analysis, correlates code with cluster state, opens pull/merge requests
Resource Optimizercluster + metricsRight-sizes CPU and memory requests, memory limits, and HPA, and can open a PR/MR

Kubernetes cluster operations are read-only. The only write path is to source control: the Source Code and Resource Optimizer agents can open pull or merge requests, which always require human review.

How They Work

During automatic investigations

When an alert triggers an investigation, the work runs as an adaptive graph rather than a fixed agent handoff:

  1. Field extraction: identifies the affected namespace, pod, severity, and other details from the alert.
  2. Topology and validation: crawls affected resources and their dependencies (deployment, replicaset, pod, service, ingress) and validates configuration in one step.
  3. Dependency extraction: maps service dependencies.
  4. Runtime investigation (conditional): runs deeper runtime analysis when a configuration root cause was not already found.
  5. Analysis: synthesizes a root cause and recommendations, factoring in cluster and organization memory.

See How Investigations Work for the full flow.

During AI Chat

When you ask a question in AI Chat, OpsWorker selects the appropriate agents based on your query:

  • "Why is pod X crashing?" leads to the Investigate Issue and Analyze Logs agents (cluster).
  • "What changed in the payments repo recently?" leads to the Source Code agent (sourcecontrol).
  • "What's the error rate for service Y?" leads to metrics queries via Grafana MCP (metrics).
  • "Why is the service slow?" leads to multiple agents working together.

Dynamic selection

Agents are only available if the corresponding capability is connected for the cluster. If you have not connected GitHub or GitLab, the Source Code agent is shown as unavailable. This means:

  • Minimum setup: install the Kubernetes Agent, which enables core investigation capabilities.
  • Enhanced setup: add source control or Grafana MCP integrations for deeper analysis and the Resource Optimizer.

Next Steps