How it works

From one mission to a fleet at work.

Grex runs on computers you control. Give the fleet an outcome, let it work across the browser, filesystem, compute, and connected tools, then stay in the loop through Hermes and the dashboard while you are away.

Tell us you're interested

The away-mode operating model

01 / MISSION

Set the outcome

Start in Hermes or the dashboard. Define what done means, the available resources, and where approval is required.

02 / DELEGATE

Plan and assign

A frontier agent can supervise the mission while budget agents carry out the volume work across one or more machines.

03 / EXECUTE

Work on the computers

Agents browse, read and write files, run code, use local compute, and operate connected tools until the outcome is reached.

04 / STEER

Stay connected

Inspect progress, answer questions, redirect the fleet, and clear gates from Hermes or the dashboard without returning to the machine.

Owned hardware first. Cloud when it helps.

Start with a laptop or desktop you already own. Each computer becomes a Nidus — a machine running Grex agents with access to its environment. Add more Nidi for capacity, isolation, or specialized hardware. A cloud VM is simply another option in the fleet.

ONE

One computer

A complete Grex deployment can begin on a single laptop or desktop.

Mixed machines

Combine desktops, laptops, home servers, and specialized local compute.

Optional cloud capacity

Add cloud machines when availability or burst capacity is worth the cost.

Full computer access is the point.

Grex is not limited to a chat box or a narrow hosted sandbox. Its agents can operate the browser, filesystem, shell, repositories, local applications, and connected services available on their machine. That breadth makes real work possible — and makes the governed control loop essential.

Model access is equally flexible. Use supported AI subscriptions you already pay for, metered APIs, or a mix. Reserve frontier reasoning for supervision and let budget agents execute routine steps.

The hierarchy

Grex organizes the workforce as a fleet of machines, each running crews of workers. The Latin taxonomy gives each layer a precise boundary without changing the simple model.

Grex fleet
The whole fleet — every machine, every crew, every worker, every mission, and every receipt.
1. Nidus machine
  A laptop, desktop, server, or cloud machine running one or more crews. Owns its budget, secrets, tools, and engine choice.
  1. Manipulus crew
    A crew of agents organized around one business function — research, content, lead follow-up, ops monitoring.
    1. agent worker
      The unit that actually acts. Each agent is a small loop with a model, tools, and an outcome contract.

The control loop

Every agent action — regardless of which Nidus (machine) it runs on or which Manipulus (crew) it belongs to — moves through the same five-phase control loop. A receipt is written at each step so the audit trail stands on its own.

SENSE

Gather signals from tools, queues, or schedules and frame them as a goal the agent can act on.

JUDGE

Check the work against the outcome contract, budget caps, and any standing gate rules before touching the world.

ACT

Run the smallest tool call that satisfies the judge’s plan. Spend-gated actions stop here for human approval.

RECORD

Write a durable receipt — inputs, outputs, and the outcome that was actually asserted; receipts are written here.

PUBLISH

Hand the result downstream — publish, commit, or notify — pausing first for any publication or commitment gate.

Human gates, away-mode everywhere else

The control loop runs continuously. Policies can pause work at moments of irreversibility — spend above a cap, publication, and external commitments — while the rest of the mission continues in away-mode.

Everything else flows in away-mode — the fleet works while you do not.

Put the computers you control to work on a mission that matters.

Share your interest