What is burrow?

Burrow is an open source tool that lets an AI coding agent, such as Claude Code, Cursor, or Codex, operate your own Kubernetes cluster to deploy, scale, debug, and roll back real applications. Rather than asking you to learn a new dashboard or adopt another agent, Burrow installs as a single binary that your existing agent can call, so you keep talking to the tool you already use while it does the actual work of running your infrastructure. The core idea is that the agent never holds your cluster credentials directly. A control plane you own keeps those credentials and enforces rules called guardrails around risky actions such as deploying to production, deleting an app, exposing something to the internet, or changing DNS. You set each guardrail to confirm, meaning it asks for your approval first, or deny, meaning it blocks the action outright, and rules can differ between environments so an agent can move quickly in staging while production stays locked down. The agent proposes an action, you decide, and the boundary is enforced in code rather than relying on the agent simply following instructions. Once connected, you operate your apps in plain language, telling the agent to deploy to production, roll back a release, scale a service up or down, or investigate why a site is slow or unreachable. Burrow adds tooling such as centralized logging and a metrics collector so the agent can read what actually happened in your app instead of guessing, then propose or apply a fix. It also supports optional add ons, including a shared Postgres database, logging, metrics, and a cache service, which the agent can request and install with your approval. Burrow describes itself as different from platforms like Vercel, which run your app on their own infrastructure. Instead, Burrow operates your own Kubernetes cluster, handling self healing workloads, rolling deploys, ingress, TLS, and upgrades, and it works with any cluster reachable through kubectl. The project is licensed under Apache 2.0, is written in Go, and is aimed at developers who want an AI agent to manage real production infrastructure without giving up control over what that agent is allowed to touch.