Debug Kubernetes
failures in seconds.
KubeIntellect finds root cause automatically, proposes the fix, and executes it once you approve — no digging through logs.
Connects through your kubeconfig
Peer-reviewed — published in the Journal of Grid Computing
Real session · no cherry-picking
Works with your existing Kubernetes stack
Interactive Demo
See it in action
Provision a namespace, deploy a web app, expose it, scale it — all in plain English.
Real session · no cherry-picking
Use Cases
Three things we do exceptionally well
Not a platform. Not a framework. A tool that solves real Kubernetes pain.
Debug failures instantly
"Why is my pod failing?"
Get root cause analysis in seconds. KubeIntellect correlates logs, events, and config to tell you exactly what broke and the exact command to fix it.
- Identifies CrashLoopBackOff, OOMKilled, ImagePullBackOff and more
- Correlates container logs with Kubernetes events automatically
- Provides a concrete kubectl fix — not just a description
Create deployments via chat
"Deploy FastAPI with Redis on K8s"
Describe what you want. Get production-ready YAML with resource limits, probes, HPA, and RBAC — no templates to hunt down.
- Generates Deployment, Service, Ingress, and HPA
- Includes liveness/readiness probes and resource limits
- Best practices applied by default — no hand-holding required
Monitor clusters in plain English
"Is my cluster healthy?"
Get a plain-English health summary. Spot resource pressure, error rate spikes, and impending evictions before they page you.
- Detects node memory pressure and CPU throttling
- Flags elevated error rates with service-level context
- Proactive suggestions to prevent incidents
Why KubeIntellect
Not kubectl. Not ChatGPT. Not both.
Pasting kubectl output into ChatGPT gets you advice. KubeIntellect gets you results — because it has direct access to your cluster.
Acts, with approval
It acts — it doesn't just explain
kubectl + ChatGPT tells you what command to run. KubeIntellect runs it: reads straight away, and writes, execs or deletes only after you approve — within the role of your key.
Parallel evidence
Specialists where they pay off
A lean coordinator answers simple questions directly. For a real fault it sends four subagents — pod state, metrics, logs, events — out at once and merges what they find into one root cause.
Peer-reviewed origins
Built on a published design
The published v1 architecture — a supervisor, 13 specialized agents and a Code Generator that wrote new tools at runtime — resolved 93% of 200 queries in its peer-reviewed evaluation. Today's v4 keeps what earned its place and is leaner.
Workflow memory
Handles multi-step tasks end-to-end
KubeIntellect remembers context across steps in a workflow. Diagnose, patch, verify, roll back — all in one conversation, without losing state between commands.
Peer-reviewed research
Not a demo. A measured system.
KubeIntellect isn’t a weekend prompt wrapper. Its architecture and results were published in the Journal of Grid Computing — evaluated on a live four-node cluster running 170 pods across 18 namespaces.
Results from the peer-reviewed evaluation of the published v1 architecture: 13 domain-specialized agents coordinated by a LangGraph supervisor, with human approval required before any write to the cluster.
How It Works
From question to fix in three steps
Fully automated. No digging through logs.
Connect a cluster
Install KubeIntellect and run the setup wizard: it takes your OpenAI or Azure OpenAI key and uses your kubeconfig — or creates a local Kind cluster, so you can try it on a laptop in minutes. Runs on your cluster with your LLM key; the Helm chart's default RBAC is read-only — no cluster-admin required.
Ask in plain English
Type your question naturally — no kubectl syntax, no PromQL, no YAML hunting. Just describe the problem. Here is how natural-language kubectl works.
Get instant diagnosis + fix
AI agents correlate logs, events, and metrics simultaneously. You get root cause, explanation, and the exact command to fix it.
Security & Setup
Connects in 30 seconds. Your cluster stays yours.
No agents to install. No cluster-admin. No credentials leaving your machine.
Uses your existing kubeconfig
No new agents to install on the cluster. The self-hosted KubeIntellect server reads the same kubeconfig kubectl uses — local file or environment variable.
No cluster-admin required
Read permissions are enough to start: the Helm chart's default ServiceAccount is get/list/watch only. Write and exec verbs are opt-in RBAC settings — nothing more than what you enable.
Human approval before any write
By default, every destructive operation — delete, scale to zero, patch — requires your explicit approval. You see the exact command before it runs.
Your data stays yours
Self-hosted, the KubeIntellect server runs on your side and your cluster credentials stay with it. Questions and the cluster evidence needed to answer them go to the LLM provider you configure.
Up and running in minutes
One pip install. That’s it.
No agents to deploy, no Helm charts, no cluster-admin. Install it, run kubeintellect init — it uses your kubeconfig or creates a local Kind cluster, with your own LLM key — then just ask: kq -q "why is my pod crashlooping?".
Running on your cluster in three commands
The setup wizard asks for your OpenAI or Azure OpenAI key, uses your kubeconfig — or creates a local Kind cluster so you can try it on a laptop — and configures the kq CLI. Read-only by default; no cluster-admin required.
pip install kubeintellect
kubeintellect init
kq -q "why is my pod crashlooping?"Python 3.12+. Docker Compose and Helm paths are in the docs.