Interfaces as projections
The Application & Agents layer is where people and machines actually touch the system. Its defining principle is that every interface is a projection of the same model — not a separate application with its own copy of the data. A workspace, an assistant, and an autonomous agent are three views onto one underlying knowledge and reasoning structure.
This is what keeps work from fragmenting. In conventional stacks, each app owns its own store, and integration means copying data between silos that immediately begin to drift. Here, there is nothing to copy: the surfaces read and write the same objects, so everything stays coherent by construction.
Three kinds of surface
The layer expresses the shared model through three complementary surfaces:
- Workspaces
- Human-driven environments where people read, author, and connect knowledge objects directly. Every edit writes back to the shared model rather than into a private document.
- Assistants
- Copilots that operate alongside a person, grounded in the same knowledge and reasoning. They cite evidence from the graph, so their suggestions are traceable rather than opaque.
- Autonomous agents
- Agents that execute multi-step work against the model under explicit decision rights and guardrails, escalating to humans where accountability requires it.
One shared context
Because humans and agents operate over the same context, their work compounds. An assistant's draft becomes a knowledge object a person can refine; an agent's action leaves an auditable trace attributed through the identity graph; a human decision becomes context the next agent respects. No surface is a dead end, and no interaction is thrown away.
The reference implementation
This layer is abstract by design — any application built on the shared model belongs here. To see it made concrete, the Institutional Intelligence Operating System is the reference implementation of the entire architecture, including this layer.