KubeIntellect is not a demo or a claim on a landing page — it is a peer-reviewed, published system. The framework is described and evaluated in KubeIntellect: A Modular LLM-Orchestrated Agent Framework for End-to-End Kubernetes Management, published in the Journal of Grid Computing (Springer), volume 24, issue 3, 2026. And it is open access: you can read the entire paper online.
Read the paper
The complete, peer-reviewed paper is open access — read the full version here:
- Published version — the complete paper (Springer, open access): doi.org/10.1007/s10723-026-09837-6
- ACM Digital Library: dl.acm.org/doi/10.1007/s10723-026-09837-6
- Earlier preprint (arXiv): arxiv.org/abs/2509.02449 — an earlier, shorter version; for the complete work use the open-access published version above.
What the paper shows
Most “AI for Kubernetes” work stops at answering questions about a cluster. KubeIntellect goes further: a supervisor coordinates a set of domain agents that can look, reason, and act on a real cluster, behind a human-in-the-loop approval gate for anything that changes state. The paper evaluates that system on a live four-node cluster running 170 pods across 18 namespaces — not a toy setup — using a corpus of 200 operational queries and a 16-scenario fault-injection corpus scored with an eight-dimension LLM-judge rubric.
The headline results
- 93% query resolution — 186 of 200 operational queries resolved end to end.
- +25 percentage points over a raw GPT-4o baseline — 75% vs 50% on a matched task set, showing that the orchestration and tools, not just the model, do the work.
- 81.8% runtime tool synthesis — on 63 of 77 genuinely novel requests, the agent wrote, validated, and ran a new tool at runtime instead of giving up.
- 7–10 seconds and about $0.036–0.039 per query — fast and inexpensive enough to sit in a real operator workflow.
Why it matters
These numbers come from a controlled, reproducible evaluation that went through peer review — which is a very different bar from a marketing benchmark. If you are deciding whether an AI system can safely help run production Kubernetes, a published methodology you can read, check, and cite is exactly what you want. The architecture, the safety design (AST validation, time-bounded execution under the ServiceAccount’s RBAC, and the human approval gate), the judge rubric, and the full result tables are all in the open-access paper.
Try it, then read it
KubeIntellect is open source. Explore the framework on GitHub, install the kube-q command-line client from its repository (or pip install kube-q), and see the results and citation on the research page. Then read the full paper.