Agmia is an open protocol that runs a whole company as one agent. Objectives go in. Every agent session works toward them, and each one's lessons become context for the next.
A model improves in its weights: months. An agent rewrites its own parts: hours. A company improves itself: minutes.
An agent has a goal, a context, actions and a reward, and it runs a loop. A company has all of these too, spread across hundreds of agent sessions that never see each other. Agmia connects the sessions and closes the loop.
We call the third level Org RSI: recursive self-improvement, at the scale of an organization.
One writer per session. Hidden wins, and the machine decides.
Spec v0 and the reference hub, AGImy, are coming under Apache-2.0.
Most tools put agents to work when a person asks. Agmia sits where the company reads its own objectives and decides the next step itself, with the agents you already use.
left to right: a person asks → the company decides · bottom to top: one agent → the whole company
Vizzy Labs runs on Agmia. For eight weeks we watch one number: the human attention it takes to move an objective forward.
illustrative dataWe are a small team at Vizzy Labs betting that the next unit of intelligence is the organization. If you want to build the loop and be measured by it, come talk to us.
Open roles at Vizzy Labs