KubeIntellect

KubeIntellect is an open-source, LLM-orchestrated multi-agent framework for Kubernetes operations. You ask a question in plain English - for example "why is my api-server pod crashlooping?" - and it fans out to specialized agents that query kubectl, Prometheus (PromQL) and Loki (LogQL) against the live cluster, correlates the evidence, and answers with a root cause instead of a wall of logs.

It does not stop at read-only diagnosis. KubeIntellect can also change the cluster - scale, restart, delete - but never silently. Every mutating operation pauses and shows the exact kubectl command it proposes to run, and nothing executes until a human approves it. Access is gated by role-based access control with admin, operator and readonly roles, so a read-only user can investigate an incident without any ability to act on it.

Main capabilities:

• Natural-language root cause analysis correlating cluster state, metrics and logs
• A coordinator that delegates to dedicated pod, metrics, logs and events subagents
• A human-in-the-loop approval gate on every destructive action, backed by RBAC
• A terminal client (kq) plus a self-hostable HTTP API server
• Works against the observability stack you already run instead of replacing it
• An optional flight recorder that hash-chains each session so it can be replayed and verified

It is aimed at platform engineers, DevOps engineers, SREs and Kubernetes operators who want conversational troubleshooting with a hard safety gate on anything that changes the cluster.

Install paths include a pip-installable CLI, a pip-installable server, a Docker image and an in-cluster deployment, and there is a hosted read-only demo you can try in the browser. It requires Python 3.12 or newer and an OpenAI or Azure OpenAI API key.

KubeIntellect is released under the AGPL-3.0 license, and its architecture is peer-reviewed and published in the Journal of Grid Computing.