All posts
Independent5 min read

Our stack. Our model.

Built and run by Thomas Conway — a solo, independent build, not a lab and not a wrapper around someone else’s answer API. Inference is served through MLX on our own machine.

§ 01

A replaceable checkpoint

cognira Retrain 12B is the live model, retrained on a Gemma 4 base. The earlier 188M Omega checkpoint is deprecated in-product, but stays published, open-source, on Hugging Face.

The checkpoint sits behind an interface, so memory, safety rules, and user ownership survive future model changes. cognira is a workshop, not a monument.

  • Inference is served on our own stack, not forwarded to another model API.
  • Checkpoint-specific thresholds live in calibration, not product code.
  • The learning loop survives future checkpoint replacements.

§ 02

Why we tell the truth

“Most AI marketing is fiction. Ours can’t afford to be — you’d notice tomorrow if we lied today.”

A model that improves in public has nowhere to hide. That is why rejections are logged at /learning, why there is no invented carbon figure, and why we still say frontier systems will beat this one on trivia.

§ 03

Who to write to

Email [email protected]. A human writes back. More on the person behind it on the about page.