Interface surface
The same habitat can be reached from multiple interfaces without splitting identity.
Umwelten frames agent work as infrastructure: habitats hold state, evals measure behavior, and session analysis closes the loop between what happened and what should happen next.
The same habitat can be reached from multiple interfaces without splitting identity.
Config, persona, tools, memories, skills, and agents form the persistent center.
Evals and session analysis reveal what the model is actually doing over time.
The environment persists: memory, tools, persona, and sub-agents all remain stable across interfaces.
The same prompt runs across providers so latency, quality, and cost become measurable rather than anecdotal.
Past work turns into searchable context and extracted learnings that can harden the next run.
Initialize the environment so the agent has a real place to exist.
Compare providers and locals on the same tasks with score, cost, and timing.
Index and learn from past work to keep the environment improving.