Persistent agency
Intelligence becomes more useful when it can remember, develop context, and carry work forward.


Independent AI Research
We build open models, local inference, persistent agents, and controlled systems— intelligence designed to remain yours.
Explore the thesisIntelligence should compound—not reset.
The dominant AI paradigm is built around discontinuity.
Every session begins again. Every model replacement abandons context. Intelligence is rented from somewhere else, measured one token at a time.
We are building the opposite: intelligence that persists across models, machines, sessions, and time.
Models should be inspectable, adaptable, and owned by the people who run them.
Capability should live where you do—on hardware you control, without a permanent meter running.
Intelligence becomes more useful when it can remember, develop context, and carry work forward.
Powerful systems need authorization, observability, evidence, and hard operational boundaries.
From model to machine, each system advances the same thesis.
Field Notes
Latest / Open Models / Release
Introducing Tess-4.5, a 27B open-weights model for real engineering work. Level with its base on BenchLocal with 25% fewer tokens, and 96 tokens a second on a single RTX 4090.