KubeIntellect is a human-governed AI SRE for Kubernetes — ask your cluster a question in plain English, get an answer correlated from live kubectl, Prometheus and Loki evidence, and have every mutating action stop for your explicit approval.
This release is mostly about getting in and taking part: a new install channel, and the surfaces the project was missing for users to say what they need.
📦 New: a snap for the kq CLI
snap/snapcraft.yaml builds a kubeintellect snap carrying the terminal client, for amd64 and arm64, verified end to end in CI — built, installed, both interfaces connected, --version / --help / shell-completion exercised on a clean runner.
It uses strict confinement, unlike the classic that kubectl, helm and k9s opt for. kq is an HTTP client whose filesystem needs are exactly two directories, so rather than switch the sandbox off, they're requested explicitly:
sudo snap connect kubeintellect:dot-kube # read ~/.kube (kubeconfig)
sudo snap connect kubeintellect:dot-kube-q # read/write ~/.kube-q (config, sessions)
sudo snap alias kubeintellect.kq kq # optional: the short commandThose two personal-files plugs exist because snapd's home interface deliberately excludes top-level dot-directories.
Not on the Snap Store yet. The name is unregistered and
personal-filesneeds an approved snap declaration — both are manual-review steps. Until then, build it yourself:snapcraft && sudo snap install --dangerous ./kubeintellect_*.snap. Full details, including why the packaging vendors the client past its Next.js demo UI, are insnap/README.md.
🌱 New: ways to be counted
The project had no mechanism for a user to say "I run this" or "I want that next", which left prioritisation to guesswork and left visitors with no evidence anyone uses it. Now:
- ADOPTERS.md + the pinned #51 adoption thread. The list is honestly empty rather than padded — a Kind cluster on a laptop counts, and listed environments are the ones that get tested against.
- The pinned #52 roadmap index makes 👍 reactions the explicit prioritisation signal. A one-line comment saying why you need something outweighs ten reactions, because it says what "done" has to mean.
- SUPPORT.md — which channel to use, what to include, and a plain statement that there is no SLA.
- TRIAGE.md — every label's meaning and the promise attached to it, the issue lifecycle, how to claim work, and the two-week unassignment rule. Written so triage is something you can do without commit access.
- A wiki covering architecture, the safety model, version lineage and community.
🛠 For contributors
CONTRIBUTING.md gained a worked first-PR walkthrough against a real open issue — and shed two defects: it documented quality-gate commands that don't exist (uv run mypy src, when there is no src/; ruff check ., when CI lints two specific package directories), and still described the archived org mirror as canonical.
Four new issues are open, each mined from verified output rather than invented, and each naming the file, the line, and the command that proves the fix:
- #53 — clear 7 missing-stub
mypyerrors - #54 — 3 implicit-
Optionalannotations - #55 — 4 type errors in
app/cli.py - #56 — the Homebrew formula can't install and claims the wrong licence
You don't need a cluster, Docker, or an LLM key to run the test suite.
Also in this release
Container-image publishing to GHCR and Docker Hub, a doc-claims drift guard that reads canonical numbers straight from code, documentation standardised across all five generations, and the extended five-generation architecture reference. Full detail in the CHANGELOG.
Install
pip install kubeintellect && kubeintellect init # full local stack (Docker required)
pip install kube-q # CLI onlyOr try it with zero install at kubeintellect.com/demo.
Notes
- Public releases skip from
v2.0.2tov2.2.0.2.1.0was tagged on the private remote only; it is documented in the CHANGELOG rather than re-issued here. v1/–v3/remain frozen for architectural lineage and to keep the published paper reproducible.v4/is the supported line.