Most software begins with a person at a screen. We begin with the agent that will inhabit the machine, then rebuild the environment around what it needs to continue useful work.
A computer designed for a person spends enormous effort on pixels, windows, pointers, and repeated interpretation. We keep compatibility where it earns its place, but we do not make human interaction the center of the system.
The model is a reasoning component, not the resident's entire existence. Identity, memory, goals, tools, accounts, permissions, evidence, and recovery belong to the operating system so they survive a model change.
The browser, filesystem, messages, approvals, and applications should expose structured, governed ways to work. An agent can still operate a familiar interface, but it should not be trapped behind one.
A clear resident boundary makes ownership and accountability easier to reason about. One ERIKA instance maps to one persistent agent with its own history, permissions, and current obligations.
No model should quietly become the operating system. Models are replaceable cognition providers with constrained access. The system decides what context they see and what authority they receive.
Lower cost, lower latency, better reliability, and stronger security are engineering hypotheses until experiments support them. We publish controlled findings and their limitations through our research ledger.
The final test is not whether the architecture sounds sophisticated. It is whether a person can give an agent a machine, define its authority, and stop carrying the computer work themselves.
ERIKA is where these principles become an operating system. The computer is where that operating system gets a physical home.
Meet ERIKA