Skip to content

CF-04 · A portfolio operator

What an unattended agent is not allowed to do is enforced by the database, not by asking nicely

15 min
dispatch cycle
6:45
daily briefing
1
switch to halt everything

The problem

Running one agent is a project. Running a dozen unattended is an operations problem: who dispatched what, did it really finish, and how do you stop everything at 2 a.m. if something goes wrong?

What we built

  1. 01A dispatcher that sweeps for work every 15 minutes
  2. 02A grader that checks returned work against a rubric before accepting it
  3. 03A morning briefing at 6:45 and an afternoon digest at 4:30, with SMS commands to steer
  4. 04A halt switch that stops the fleet and kills running workers

What to take from it

The prompt is not the security model. Permissions are.

More on custom ai agents

Tell us what is eating your week

A free discovery call with a solutions architect, or a blueprint in about a minute.

Call usBook a discovery call