Documentation

Kopilot is an approval-gated AI agent for Kubernetes operations. It is a Python application built on LangChain and LangGraph: a supervisor agent routes requests to skill-specific sub-agents, and every command those agents want to run passes through a safety-checked executor.

Where to start

• Quickstart — install from source, configure a provider, run your first task.
• Concepts — how the supervisor, skills, executor, and operator fit together.
• Security & safety — the approval workflow, what is blocked, and what the safety layer does not protect against.

Reference

• API & CLI — every REST endpoint, request/response shapes, and CLI commands.
• Configuration — all environment variables with their defaults, generated from the settings classes in the source.
• Deployment — Helm chart, container image, quickstart manifest, and RBAC notes.
• Skills — the six built-in skills and how to write your own in YAML.

Guides

• Cost optimization — an end-to-end illustrative session, and what Kopilot deliberately refuses to do automatically.
• Interoperability — MCP server, portable skill manifests, and the agent discovery manifest.