Introduction to Quox

Put AI agents to work on your own systems. Keep a receipt of what they did.

Maturity: Beta overall. This page is an overview. Each capability links to its own page, and each of those pages states its own maturity.

At a glance
What it is	Quox is a control plane for AI agents. Agents get real access to your systems, and you decide what they may do, approve what matters, and keep a record of every governed action.
Where it runs	QuoxCORE runs on your own server, in Docker. The `quox` command line talks to it from any machine.
When to reach for it	You want AI agents doing real operational work, and you need to say yes before anything risky happens and prove afterwards what was done.
How to start	`curl -fsSL https://get.quox.ai \| sh`, then follow Install and deploy.

The five ideas

Idea	What it means	Where to try it
Agents	Specialist AI assistants, each with its own job and toolset, coordinated by CommanderQ. QuoxCORE ships agent definitions for security, infrastructure, networking, monitoring, research and more.	`quox chat`, or the dashboard's chat
Tools	The actions an agent can take: read a host, run a command, call an integration, start a workflow. An agent can only use the tools it is granted.	Agents
Approvals	Risky actions pause and ask a human. The agent proposes; a person approves or rejects; only then does it run.	`quox approvals list`
Safety levels	An instance-wide level from GREEN to RED. ORANGE sends autonomous work to a human instead of running it. RED halts new autonomous work and stops running loops.	`quox safety status`, `quox safety defcon`
Evidence	Governed actions are recorded as signed envelopes in a hash-chained log, so the record cannot be quietly edited later. You can download a run's evidence and check it yourself.	`quox volt fetch <runId>`, then `quox volt verify <file>`

How a request flows

You ask, in chat or with `quox chat`
  -> CommanderQ picks the agent for the job
  -> the agent chooses a tool it is granted
  -> policy and the safety level are checked
  -> risky? it waits for your approval
  -> the tool runs: on this server, or on a host through QuoxAgent
  -> the action is recorded as evidence
  -> you get the answer

Hosts you want agents to work on run QuoxAgent, a small per-host daemon. Without it, agents work on the QuoxCORE server and its integrations only.

What is in the box

QuoxCORE installs as eleven containers on one server. The ones you will meet first:

Part	Job
Dashboard	The web app, over HTTPS on port 3000
Collector	Chat, agents, tools, fleet and AI routing
Auth	Accounts, organisations, roles
Memory and Qortex	What agents remember, and the knowledge layer they search
QuoxFlow	The workflow engine

The full list, with ports, is in Install and deploy.

You bring the model

Quox does not include an AI model. You connect the provider your organisation already uses, with a key or a self-hosted server:

bash

quox provider connect anthropic --key-from-stdin
quox provider connect ollama http://<host-lan-ip>:11434

Supported kinds in CLI 0.3.1: openai, anthropic, ollama, openrouter, huggingface, groq, together, lmstudio, vllm and custom.

Limits

This page is an overview, not proof. The five ideas above are built and ship in QuoxCORE. How far each has been driven live is stated on its own page.
A fresh install has not yet been driven to a first chat answer. The install itself is proven; the first-answer step is next. See Install and deploy.
No model is included. Agents do nothing until you connect a provider.

Install and deploy: get your own instance running.
Agents: the specialist agents and how they are routed.
Safety: approvals, safety levels and the emergency stop.
Evidence spine: how actions become verifiable records.
QuoxCLI: every quox command.

Last verified 2026-10-29 against QuoxCORE 1b8bf54d and CLI 0.3.1 (code), plus the install run of 2026-10-27.