Get started
ArchitectureQuoxCOREDashboardEvidence

Two front doors, one platform

Adam Cowles8 November 20267 min read
Two small glowing doorways, one narrow and blue, one wider and violet, standing on a vast layered platform of lit grid planes

Small on the surface

Quox looks small when you first meet it. One install command. A terminal where you type what you want in plain English. A dashboard in your browser.

Underneath is a whole platform: 63 agents in 10 families, a store of 52 products, and a control layer that checks every request before a model runs and writes every tool call to an audit record. You do not have to learn any of that to get work done. It is there when you want to look.

The surface stays small on purpose. Most people have one way they like to work, and a tool that forces the other way on them loses them within a week. So QuoxCORE has two front doors, and both open onto the same platform, the same data and the same rules.

Two front doors

The first door is the terminal. quoxcli 0.5.0, the released CLI, has 137 top-level commands and 828 documented commands in the command reference, as of 8 November 2026. You rarely need to learn them.

Type quox with nothing after it and you are in a conversation: say what you want, and Quox hands it to the right agent, shows each step as it happens, and waits for your yes before anything that needs approval.

The second door is the dashboard. It is a full browser workspace, with its own views for your mail, your SEO, your security work and your servers.

Some people will live in the CLI and never open the dashboard. Others will run their week from the dashboard and never type quox. Both are first-class. Every capability ships with a quox verb, so the terminal reaches all of it. The dashboard covers the day-to-day work in the areas below, with the same records behind it.

A morning from the terminal

It is Friday. The inbox filled up while you were out, and you have ten minutes before your first call. You type one sentence: what in my inbox actually needs me today, ignore the newsletters and anything suspicious, draft a reply to the most urgent one and do not send anything.

In the recorded inbox run, Quox named the three people waiting on a reply (a supplier, a customer and a colleague), with the reason for each. It set aside the newsletters, the automated notices and a look-alike "PayPaI" phishing mail, and told you not to click its link.

It drafted a reply to the supplier whose delivery slot was about to go and queued it for approval. Nothing was sent. The whole job took 64 seconds and five tool calls, and the mailbox was a fictional demo fixture.

Next question: which of my dev servers are behind on the Quox skill pack, and are the two I care about ready? Read only, change nothing.

In the recorded dev servers run, Quox read the record for all six registered servers in 27.7 seconds without logging into any of them. It found the one server that was two versions behind and had not reported in for 24 days, named the one sync command worth running, and ran nothing.

That is the shape of the terminal door. You ask in your own words, Quox picks the tools, and you see each one run. In one recorded run, 16 of 318 tools were offered for the job, so the model works from a short, relevant list rather than everything at once.

The same morning from the dashboard

Now take the same morning without a terminal.

InboxQ has already filed your mail as it arrived into Focus, FYI, Auto and Screened. You open Focus and the people waiting on you are there. The reader shows SPF, DKIM and DMARC results and phishing badges on each message. Block a sender once and the next message from them is screened on sight.

Anything the InboxQ agent proposes, a draft reply included, waits for a person to approve it.

The Dev Servers board shows each enrolled server with its capability profile, its skill pack and a pill per capability. It reads the same server records the terminal question did, so both doors give you the same answer about which server is behind.

Neither door is a lesser copy of the other. They are two ways into the same records.

Four kinds of work, either door

Inboxes and servers are two of the jobs. Four published customer journeys show the range, each one a real recorded run that a machine checked against the CLI, with a downloadable evidence bundle you can verify yourself.

Busy inboxes. Who is waiting on you, in order, and the first reply already written. The inbox journey is above.

Several projects across several servers. Which servers are behind and whether the ones you care about are ready, without a login on any of them. The dev servers journey is above.

SEO. Traffic has gone soft before a marketing review. In the SEO journey, Quox read the latest crawl, found 201 indexable pages that nothing on the site links to, ruled out accidental noindex, traced which pages lost text and when, and confirmed the crawl history was intact across 134 crawls.

It handed the ranking work to its SEO specialist agent, then named one page to fix first and exactly what to change. It edited nothing. In the dashboard, QuoxSEO carries the same data as overview, changes and opportunities tabs.

Security. A change window tomorrow, and an auditor who wants proof. In the security journey, Quox handed the job to ARGUS, its Wazuh specialist, which confirmed both servers were connected and ranked 1,310 vulnerabilities by whether attackers are actually using them. The two on CISA's exploited-in-the-wild list went to the top. Two CVSS 9.8 findings stayed off the top because nothing shows them being exploited.

Then it produced a signed posture report an auditor can verify offline against a key you give them. Nothing on the servers was changed. On the dashboard side, the Quox Security console runs scoped engagements with a risk ceiling, verified targets and findings that must survive a second look.

What sits underneath

Both doors lead to the same machinery, and the machinery is the point.

Agents. CommanderQ takes the request and routes it to a specialist when one fits, as it did with the SEO and Wazuh agents above. The full roster is on the agents page.

Approvals. Reads run. Writes that matter wait for a person. In the inbox run, the only write was queuing a draft, and the draft itself was a proposal waiting for a yes.

Evidence. Every tool call in every run above was written to the audit record as an evidence envelope, and the request passed identity and policy checks before any model ran. Every CLI guide's "Ask Quox" conversation is a recorded, machine-checked run, linked to its test report. If you want to see what happened, there is a record of it, not a summary someone wrote afterwards.

Products. The store lists 52 products that add to the same platform, InboxQ among them.

How to start

QuoxCORE runs on your own Linux server or VM with Docker Compose. The install page has the one command, and two other routes in. The CLI has a guided setup: when it finds no QuoxCORE, it offers to set one up on that machine in Docker, brings the services up and signs you in. The first install builds from source and takes tens of minutes.

Or point Claude Code at Quoxpert and let it run the same install for you.

Then pick your door. Start with the CLI guide and ask a read-only question, or open the dashboard and look at what is already there. Either way you are on the same platform, and you can walk through the other door whenever you like.