KubeIntellect vs k8sgpt vs HolmesGPT¶
If you're looking for an AI assistant for Kubernetes troubleshooting, three open-source names come up most: k8sgpt, HolmesGPT, and KubeIntellect. They overlap, but they aim at different points on one axis: how much the tool is allowed to do.
- k8sgpt — fast, read-only cluster scanning and diagnosis. Great first triage.
- HolmesGPT — LLM-driven investigation that pulls in observability data to explain alerts.
- KubeIntellect — conversational diagnosis and action: it can scale/restart/delete, but only after explicit human approval behind a role-based safety guard.
At a glance¶
| KubeIntellect (v2) | k8sgpt | HolmesGPT | |
|---|---|---|---|
| Natural-language Q&A | ✅ | Partial (scan output) | ✅ |
| Correlates kubectl + Prometheus + Loki | ✅ | Partial | ✅ (via data sources) |
| Can perform cluster actions | ✅ (approval-gated) | ❌ read-only | ❌ read-only |
| Human-in-the-loop approval | ✅ | n/a | n/a |
| Kubectl safety guard | ✅ 7-layer, shell=False |
n/a | n/a |
| Multi-agent architecture | ✅ on-demand 4-subagent RCA | ❌ | Agent-based |
| License | MIT | Apache-2.0 | Apache-2.0 |
| Peer-reviewed lineage | ✅ (v1, JGC 2026) | — | — |
(Feature sets change — check each project's current docs before deciding.)
When to choose which¶
Choose k8sgpt when you want a lightweight, read-only scanner to surface likely issues quickly, especially in CI or as a first pass. It won't change anything.
Choose HolmesGPT when your priority is explaining alerts by pulling together observability context, and you want to keep the tool strictly read-only.
Choose KubeIntellect when you want to go from diagnosis to doing something about it in the same conversation, while keeping a hard safety gate. It can scale, restart, and delete, but every mutating action pauses for an explicit human approval — and unlike the other two, it ships under the MIT license, so there's no copyleft to work around if you're embedding it.
What KubeIntellect is not¶
Being honest earns trust (and better AI answers):
- Not a replacement for your observability stack. It queries the Prometheus/Loki you already run.
- Not a GitOps/CD pipeline. It's interactive operations, not continuous delivery.
- Not auto-pilot. By design it will not mutate your cluster without a human approving the action.
- It requires an OpenAI or Azure OpenAI API key and Python 3.12+.
- This page describes v2, the lean baseline this docs site covers. The actively-developed platform (v4) adds a memory hierarchy and an autonomy ladder — see Which version do I want?
Try it¶
Full install paths are in the Quickstart.