AI Agents
OpsWorker uses specialized AI agents that each understand a specific data source. During investigations and free-form AI Chat sessions, these agents gather context and, in the case of source control, can propose changes.
Each agent is gated by one or more capability tokens: cluster (the in-cluster Kubernetes Agent), sourcecontrol (GitHub OR GitLab), and metrics (Grafana MCP). An agent that is missing a required capability is shown as unavailable with its missing capabilities listed.
Free-Form AI Chat Agent Registry
Free-form AI Chat exposes six agents:
| Agent | What it does | Required capability |
|---|---|---|
Investigate Issue (investigate) | Runs a focused investigation on a problem you describe | cluster |
Validate Resources (validate_resources) | Checks configuration correctness (selectors, ports, limits) | cluster |
Check Dependencies (check_dependencies) | Examines service-to-service dependencies and health | cluster |
Analyze Logs (analyze_logs) | Searches and interprets logs for error patterns | cluster |
Source Code & Repository Agent (source_code) | Correlates code changes and opens PRs/MRs with fixes | sourcecontrol (GitHub OR GitLab) |
Resource Optimizer (resource_optimizer) | Right-sizes CPU/memory requests, memory limits, and HPA, and can open a PR/MR | cluster + metrics |
These chat agents are distinct from the investigation graph nodes (static_extraction, topology_validation, dependency_extraction, investigation, analysis). See How Investigations Work.
Resource Optimizer
The Resource Optimizer uses live utilization metrics (via the metrics capability / Grafana MCP) to recommend right-sized CPU requests, memory requests, memory limits, and HPA settings. It never recommends a CPU limit (it may recommend removing one), and it no-ops when current settings are already within roughly 15% of its recommendation. After you confirm, it can open a GitHub PR or GitLab MR with the manifest change (a draft if confidence is not high) for human review.
☸️ Kubernetes Agent
Read-only queries for pod status, events, logs, configurations, and resource topology via the in-cluster agent (core types built-in, broader types via Kubernetes MCP).
Read more →🔀 Source Code & Repository Agent
Correlates commits and PRs with incidents and can open human-reviewed pull/merge requests with fixes (the only write path in OpsWorker).
Read more →📈 Metrics & Grafana Capability
Grafana MCP tools for PromQL metrics, LogQL log search, dashboards, and alert rules. Consumed mainly by Resource Optimizer and to enrich investigations.
Read more →📐 Resource Optimizer
Right-sizes CPU/memory requests, memory limits, and HPA from live metrics, then opens a PR/MR after your confirmation. Never sets a CPU limit.
Read more →