How Egos works — the agent model

Egos is the agent operating system behind fbuild. Rather than a single assistant, it coordinates a team of agents that work within a defined scope.

Agents, units & ranks #

• Egos can spin up many agents that work like a coordinated crew — each takes on part of a task, and the one that produces the best result effectively "ranks up".
• Each agent has a unit (for example, build vs. review), a rank, and its own memory.
• What an agent uses — which model, what it keeps in memory, which capabilities it has — is configurable and can be refined over time.

Layered memory #

Egos keeps memory in layers that grow like a tree. A lower layer only talks to the layer directly above it, so only the useful, distilled information rises to higher layers instead of everything being dumped upward. This keeps context focused as work scales.

Communication by signals #

Agents don't chat freely with each other. They coordinate through a signal mechanism (notify / signal). This keeps coordination auditable and prevents agents from talking past their scope.

Scope & control #

Every agent operates within the scope you give it. You stay in control of what work is in bounds — the agents organize themselves to get it done, but they don't wander outside the boundary you set.

This article describes the model Egos is built around. Some capabilities are still being expanded — see Where fbuild is going for what's on the roadmap.