One box is governed.
A fleet of them should be too.
Qluster is the console for the QuoxCORE installations you already run. A box for research, a box for security, a box per client. One pane of glass over all of them, and not one of them gives up its autonomy to be there.
In plain words
What it is, where it lives, when to reach for it
- What is it
- A server with a web console that watches every QuoxCORE box you run, from one screen.
- Where do I use it
- In your browser, pointed at the Qluster controller you host on your own machine.
- When would I use it
- When you run more than one Quox box and want health, drift and backups in one place.
- How do I use it
- From a checkout of the repository, run
docker compose up -d; a public package is not published yet.
QuoxCORE is the free, self-hosted platform underneath this. What is QuoxCORE
Running one Quox box is a solved problem. Running nine of them is a spreadsheet, a folder of SSH configs and a quiet hope that the backups are still working. Qluster is the answer to the ninth box.
Split your boxes on purpose
Isolation is a design decision, not an accident. Give each job its own box and the damage from any one of them stops at its edge. Qluster is what makes that affordable to operate.
RESEARCH-01
Long-running analysis that you want nowhere near production data. Its own box, its own budget, its own blast radius.
Six boxes, one console. The isolation is the point: a bad deploy, a leaked credential or a runaway job stays on the box it started on.
Every box, and everything each one did
Four surfaces of the real controller. The fleet at a glance, one box in depth, the evidence trail underneath it, and what happens when you ask a box to do something.
- 1Fleet healthOne number for the whole estate. Muted boxes are excluded so it never lies to you.
- 2Node treeEvery registered box, grouped by kind. This demo fleet is one group of five, sorted worst-status-first.
- 3Per-box truthStatus, version and last-seen, refreshed live. Reachable and unreachable are different states.
- 4Live resource gaugesCPU, memory and disk sampled from the box itself, refreshed as it reports.
Captured from a running Qluster controller against a demo fleet, not a mockup. Five boxes, four online and one degraded, because that is the fleet these screenshots were taken from.
A console, not a leash
Qluster is an overlay. It does not run inside your Quox installs and it does not take custody of them. Ask a box to do something and the request still lands in that box's own policy engine, its own approval queue, its own evidence chain. It can refuse you, and the refusal is recorded like anything else.
Qluster being down never means a Quox box is down.
- 01You ask
An operator delegates a task from the console, signed and scoped.
- 02You approve
Always-on HITL at the controller. Nothing leaves on a timer.
- 03The box decides
It applies its own policy and its own approval, independently of you.
- 04Both sides record it
One correlation id, two evidence chains, verifiable after the fact.
The honest index
Every capability, tagged with how much live mileage it has. Taken from the maturity map in the repository, which we keep ahead of this page rather than behind it.
- Fleet registry and health pollerStable
- Server-truth registry with a jittered poller. Reachable and unreachable are separate states, so a service falling over does not read as a dead box.
- Datacenter view and status wallStable
- Node tree, fleet grid, RAG status tiles. Muted boxes are excluded from fleet health rather than quietly counted.
- Metrics and telemetryStable
- CPU, memory and disk gauges, 24h charts, 14-day availability. A box that does not report says so instead of showing a zero.
- Cross-instance checks and driftStable
- Version drift, config fingerprint, backup freshness, and a real remote WARD chain verify that recomputes hashes rather than trusting the box. Verification is checkpoint-anchored: it re-checks the entry it last verified, so a box that rewrites its history and re-links the chain is caught.
- Secure enrollmentStable
- Single-use tokens, scoped bearers, Ed25519-signed operational requests with replay protection, revocation inside a minute.
- Remote encrypted backup pullStable
- Trigger, stream, verify against the box manifest, then encrypt at rest. One retention policy instead of per-box scripts.
- Colonel Qluster fleet agentStable
- A read-only meta-agent over the whole fleet. Answers questions no single box can, because no single box can see the others. Runs on your own Claude session, with no API key required.
- Governed delegationStable
- Task a box from the console under always-on approval. The box then applies its own policy, so it can still refuse you. Soaked over consecutive governed round trips, each one leaving a full cross-box evidence chain behind it.
- Alerting and escalationStable
- Unreachable and recovery alerts with per-box loudness. An alert left open escalates to a separate on-call route, and acknowledging it records who owns it. Acknowledge is deliberately not resolve: only a real recovery clears an alert, so the evidence trail cannot be closed by hand.
- Cross-box evidence displayStable
- One correlation id replayed across controller and box in causal order, with the WARD entries and the VOLT bundle. Says so plainly when a box cannot be reached, instead of passing off half a timeline as the whole story.
- Evidence log and exportStable
- Every envelope with its governance verdicts, filterable by intent, box, actor, layer and time, and exportable as NDJSON or CSV. Your evidence is not trapped in our console.
- Multi-operator access controlStable
- Admin, operator and viewer against an explicit capability matrix. Disabling someone revokes their live sessions and keeps their history, so past actions stay attributable.
- Terminal and CI control surfaceStable
- A qluster CLI covering everything the console does, JSON on every command and exit codes meant for cron. Verification of a fleet can be a scheduled job rather than a habit.
Install Qluster
Deploy the control plane, then enroll the boxes you already run.
# Clone the repository
git clone https://github.com/quoxai/quox-qluster.git
cd quox-qluster
# Configure environment
cp .env.example .env
# Start the control plane
docker compose up -d# Everything the console does, and exit codes for cron
qluster status
qluster checks run # exits 2 if any check fails
# Verify a box's hash chain on a schedule, not a hunch
qluster qlogs verify --chain nw-prod-01
# Take your evidence with you
qluster qlogs export --format csv > evidence.csvLAN posture today
Qluster now supports multiple operators with roles, but it is still built for a trusted network. Public exposure waits on the hardening stream, so run it inside your perimeter for now. Source is at github.com/quoxai/quox-qluster.
Bring your fleet under one roof
Qluster works with any number of QuoxCORE boxes. One for every job, one console to run them, all governed under the wider Quox enterprise platform.
Stay updated
Product updates, new features, and protocol news.