Zaloguj sięUmów spotkanie
The AI workforce

AI agents with an owner, a limit and an off switch.

Every AI agent on Nexus has a named person in charge of it. It starts no higher than "ask first" and earns more freedom over time, the way a new hire does. One switch stops a whole fleet.

A named owner for every agentFreedom is earnedOne switch stops a fleet
A floor of AI agent workstations, one paused, with a large stop switch in front
Four levels of freedom

No agent starts above "ask first". It earns anything more.

Its owner can start it lower, at watching or suggesting, for new or sensitive work. It moves up only when its record says it should, and only when a person grants it. Your rules set the highest level each kind of job may ever reach.

  1. 01 · Watches and reports

    Reads and flags. Changes nothing. Good for new kinds of work and anything sensitive.

  2. 02 · Suggests, a person acts

    Drafts the reply, the decision or the change. A person sends it.

  3. 03 · Asks first, then acts The usual start

    Does the work once a person has said yes to the plan. The default for a new agent, and the highest it can start at.

  4. 04 · Acts, then reports

    Earned over time for routine work with a clean record. Everything it does is still recorded and can be undone.

workforce · autonomy ceilingsSample data
The highest autonomy level allowed for each kind of job
Autonomy ceilings. The most freedom each kind of job may ever have.Sample data
The floor

See every agent at work, as it happens.

The floor shows each agent, what it is doing, who owns it and what it has spent. Open any run to see each step and the gate it passed.

workforce · a run, liveSample data
One AI agent's run, live, step by step
workforce · the floorSample data
The workforce floor with every AI agent, its owner and its status
Limits and switches

Hard limits. And a switch that works.

Limits are checked before an agent acts, not after. If something goes wrong, the people in charge can stop one agent, a team or the whole fleet at once.

workforce · limits and tripwiresSample data
Limits and tripwires that stop an agent before it acts
Limits and tripwires. Money, data and time limits checked before each step.Sample data
workforce · kill switchesSample data
Kill switches for one agent, a team or the whole fleet
Kill switches. Stop one agent, a team or everything.Sample data
Who watches

A second pair of eyes from a different family.

A watcher built on a different family of models reads every agent's work against its rules. It catches shortcuts, faked work and agents that start working together in ways nobody asked for.

  • Passports

    Every message between agents says who sent it and what it may do.

  • Sentinel

    Watches for fraud patterns, attacks and agents going wrong, and raises findings for a person.

  • Honest review

    Answers that matter are reviewed by a second model before a person sees them.

workforce · passports, liveSample data
Passports as messages flow between agents
Passports, live. Each message checked as it flows.Sample data
workforce · sentinelSample data
Sentinel at a glance, with open findings
Sentinel. Findings for a person to look at.Sample data
Rooms and rulings

When an agent reaches its limit, a person rules.

An agent that needs more than it is allowed can convene a decision room. AI argues both sides, a named person rules, and the ruling is sealed with the rest of the record.

workforce · the rulingSample data
A ruling in a decision room, with reasons and signatures
The ruling. Reasons, signature and seal.Sample data
Book a briefing

Hire your first AI agent the careful way.

Pick one job your team does every week. We will show it running at "ask first", with an owner, a budget and a switch.