Deployment and integration
We put an agent in front of the machines that have nowhere to put one.
Video out of the HDMI port, keyboard and mouse in through USB, nothing installed on the target. Getting that working on your hardware is the job: choosing the path, preparing the board, surviving the target's USB enumeration, and proving the loop is quick enough to leave running. That is what we do.
High Performance General Purpose Agent Machine Interface
The complete picture in one screen.
The target sees a monitor and a keyboard and cannot tell the difference, so every screen a person can operate becomes operable, including the ones that exist because no software is running yet.
USB HID reports written straight to the gadget device, so
chords, hold, ctrl-drag and shift-click all work rather than one key at a
time.
One authenticated JSON API, bearer key, no localhost
exemption.
60 fps at 1080p and a 42 ms median action-to-visible on our
current bench, roughly six times faster than a person. Caveats under What we
measure.
The loop
Next
Where an API stopsThe jobs that stop because there is nowhere to put a program.Use casesThe eight shapes of job this reaches, and where it is the wrong tool.What to expectWhat it does well, what it cannot do, and what surprises people.Agent costsWhat an agent actually costs to run, and how the design controls it.SecurityThe threat model, stated first, and where the boundary really sits.What we doThe six steps of an engagement, from assessment to handover.What we measureFrame rate and action-to-visible, and what the numbers do not prove.Running it yourselfBuy the parts instead. Here is what you take on.