Build your agent · live

Describe the agent. Watch the factory build it.

The HexaBox Agent Factory is itself the first HexaBox agent, and the one that built Eihwaz, our AI-driven cyber-security fleet. Here you watch it construct an agent's structure: the intelligence of your choice inside, the gate at every action that touches the real world, and a seat for the person at the wheel. Describe the job in the box below; about a minute later you see the agent's structure: the top of its Studio view, its topology in 3D, and the facts of every part — what it does, what goes in and out, which authority an action needs.

0 / 1 700 characters
Try one
Help me write it
Plain words are enough. The factory sorts the job, picks the rules and the levels — you describe, it builds.

This is the real factory. Your description goes through the same chain the factory runs for a partner — sorted, split into steps, built from proven parts, checked against our build rules before it is released. In about a minute you see the agent's structure, its rules and its approval requirements. Customer implementation and deployment follow — your team confirms the customer's requirements and policy there.

The factory is working
Built · the top of the Studio view

—

This is the agent's structure. The next steps are implementation and deployment: connect it to a customer's systems, confirm their requirements, validate, and take it into operation within the agreed scope. Or bring an agent you already run — we place the controls around it.

What you just saw

Eight steps, one way in, nothing ships unchecked.

The factory's own steps are the ones you watched. Every one of them leaves an audit event; together they form the chain the agent carries for life.

01
The job comes in — one entranceYour description is the only input. No template, no questionnaire.
02
Sort the job, split it into stepsA model reads the job and drafts the steps and the competences it needs. Everything it drafts is checked downstream; a model never gets the last word.
03
Pick the rules — fixed rules, no modelThe sector and its rulebook are derived from the job, never guessed from a keyword. The data-sharing floor, the policy additions and the authority levels follow.
04
Build the map from proven partsDoors in and out, the competences, the decision, the safety check, the action selector and one approval gate per declared action — in the order the library prescribes. The same job gives the same map.
05
Write a contract for every portWhat may come in, what may go out, what is forbidden — one contract per port of every part, ~100 for a typical agent.
06
Check the 19 rules — a hard stopOne rule fails: back to the map, at most three times. All pass: a score out of 100 and a verdict that is written to the record.
07
Protect shared dataThe controls for sharing across customers are built in: counts only, never names or content, and only when at least five customers look alike. If fewer do, nothing is shared and the record says why.
08
Sign and shipA signed identity, a hash chain of every build event, and a bundle that can prove it is unchanged with one command. The demo stops before the bundle — that part is a conversation.

The demo builds at most three agents per visitor per ten minutes and rests after its daily share. Descriptions are processed by the factory and its model provider under a data-processing agreement; the page stores nothing about you unless you choose to send it to us. Details in Trust & privacy.

Next

From this structure to a running agent.

Demonstration — you have just seen the generated structure, its rules and its approval requirements. Customer implementation — the agent is connected to the customer's systems and the relevant requirements are confirmed. Deployment — validated and taken into operation within the agreed scope. The demo keeps the description you wrote so you do not have to write it twice.