Skip to main content

An agent for the job your team repeats every week.

In design. A custom agent is a company's own digital colleague for one recurring job, briefed like a new clerk. It reads, matches, chases and drafts; people approve anything that changes the records.

The problem

Some work is too varied for a fixed automation and too repetitive for a person.

  • Certificates in the inbox

    Suppliers e-mail documents that someone has to read and attach to the right lots.

  • Evidence before departure

    Every EU-bound lot should have its EUDR evidence a week before it leaves.

  • Certificates about to expire

    Suppliers whose certificates expire need a reminder, in time.

What it does

The design sets what an agent may do and how it earns trust before it is used.

  • A playbook, not code

    The goal, the steps and your own rules and exceptions, drawn on the same canvas as automations.

  • Started like an automation

    When something happens and the conditions hold, an automation runs the agent. It can also be run by hand or from an inbound e-mail.

  • A small set of tools

    Read, prepare, communicate and propose. Never approve, never administer.

  • Outside text read safely

    Inbound e-mail and documents go first through a reader without tools; the agent sees checked fields, never the raw text.

  • Tested, tried, then in use

    Scenario and adversarial tests, then at least 7 days on trial where nothing leaves, then an owner or administrator puts it in use.

  • Read the full module

How it works

From a draft in the sandbox to an agent in use, measured.

  1. Write the playbook

    The job as you would brief a new clerk.

  2. Grant its tools

    Only what the job needs, from the platform's catalogue.

  3. Test and try it

    Tests must pass, then a trial of at least 7 days in which every would-be action is shown.

  4. Measure it

    Accepted, edited or rejected work, minutes saved against a declared baseline, cycle time and cost.

Fictitious demo data

Who uses it

Corrections people make become proposed lessons; a person accepts them into a new playbook version, which is tested again.

  • Operator

    Drafts and tests playbooks in the sandbox.

  • Owner or administrator

    Puts an agent in use after its trial.

  • Approver

    Approves what the agent proposes.

  • Organization

    Publishes templates and sets policy for its companies.

Tell us the job you would hand to an agent.

Custom agents are in design. Tell us the recurring work you would brief one on, and we will walk you through the design.