Open-source Kubernetes AI assistant
Stop reading logs. Ask why.
hexawyn learns from your incidents and tells you the root cause in seconds — reliable answers backed by a deterministic engine, not just a chatty model.
$hexawyn "why is payments-api crashing?"
OOMKilled. Your JVM has no -Xmx flag, so it takes 80% of
the container (410Mi) but your limit is 512Mi.
Fix: add -Xmx384m to your deployment.
resolved in 12 seconds
3 a.m. The pager goes off.
Every incident is an archaeology dig. hexawyn hands you the answer, not another dashboard to read.
200 lines of kubectl describe.
5,000 lines of logs across three pods.
45 minutes later: it was an OOMKill.
Why the answers are trustworthy
hexawyn is 80% deterministic code, 20% LLM. The LLM is the polish, not the engine.
Collect & store (K8s, OpenTelemetry, Prometheus)
deterministic
Rules, thresholds & calculations turned into facts
deterministic
Long-term incident memory — cosine search in an encrypted local DuckDB on your machine
deterministic
Final reformulation (LangGraph)
LLM
The LLM talks. The agent verifies.
Your clusters are too critical for guesswork. Every answer is checked by a dedicated validation agent before anything happens. Dangerous commands are blocked. Inconsistencies are flagged. Edge cases are handled, not forgotten.
- Two layers of intelligence, one mission: zero hallucination.
- A four-eyes principle for your Kubernetes assistant.
- Reasoning is verified before any action is proposed.
Local-first by design
Your kubeconfig and raw logs never leave your machine. hexawyn keeps a long-term memory of your incidents in an encrypted local DuckDB file (AES-GCM, keyed to your kubeconfig) — on your own machine, under ~/.hexawyn.
- kubeconfig never leaves your machine
- Raw logs stay inside your cluster
- Long-term incident memory stored locally, encrypted at rest — no central database
Full transparency: your question and its context transit through the control-plane. Your credentials and raw logs do not.
One assistant, every part of the job
From the 3 a.m. incident to the quarterly board review — hexawyn covers the work that slows platform teams down.
Debug incidents
Ask why a workload is failing and get the root cause with a concrete fix.
Reduce cloud costs
Surface over-provisioned resources and see exactly what to right-size.
Improve reliability
Turn observability into actionability across every cluster you run.
Secure your platform
Spot risky configurations and dangerous commands before they ship.
Accelerate GitOps
Understand drift and change impact without spelunking through YAML.
Automate reporting
Weekly reliability digests sent to the people who need them.
Manage multiple clusters
One assistant across AWS, GCP, Azure and on-prem — no lock-in.
Speak the business language
Translate infrastructure into availability, cost and risk for leadership.
Works with what you already run
From vanilla Kubernetes to EKS, AKS, GKE and OpenShift — plus the observability, FinOps, GitOps and security tooling around them. If it speaks OpenTelemetry, hexawyn understands it.
A production-grade assistant, not a POC
Built on hexagonal architecture and a decade of software-engineering discipline. Not vibe coding — every line is tested.
On the roadmap
Diagnosis works today. Prediction gets sharper as hexawyn learns from your incidents.
- Predict saturation and failures before they reach your users
- Budget alerts when autoscaling is about to spike your cloud bill
- Executive reports that answer leadership questions automatically
Community Membership
Join our growing community of users and contributors. We welcome everyone interested in Kubernetes and AI.
Learn about membershipTalk to the team
A question about a plan, a self-hosted deployment or an incident you want to throw at hexawyn? Drop us a line and we'll get back to you.
Your next outage shouldn't be your first warning.
Install hexawyn, point it at a cluster, and ask your first question. It runs on your machine, in about a minute.