A research and engineering practice oriented toward the long term.

We treat software the way a machinist treats a tolerance: with patience, measurement, and respect for the material. Systems that will run for decades cannot be improvised.

Each release compounds on the last. We prefer fewer, better primitives — assembled deliberately, refined continuously, and documented as if the next engineer to read the code is a stranger.

Our models are designed to be legible. What they observe, what they infer, and why they act must be inspectable by the humans they serve. Opacity is a design failure, not a technical shortcut.

We build with explicit boundaries: retention windows, refusal semantics, and audit trails are part of the model, not an afterthought layered on top of it.

Centralized inference concentrates leverage in a handful of operators. We think that is a civilizational risk. Our infrastructure treats the edge — laptops, phones, small servers — as first-class citizens.

Local-first is not nostalgia. It is a design choice that keeps individuals, communities, and institutions in charge of the intelligence they rely on.

The tools we build are meant to give humans more range, not less relevance. We measure success in the capabilities we hand back to the person at the keyboard.

A system that removes judgement from the loop has failed our brief, regardless of its benchmarks.

The problems worth working on unfold over generations: climate, energy, coordination, knowledge. Our work is oriented toward the infrastructure those problems demand — durable, transparent, and quietly optimistic.

This is what we mean by solarpunk. Not an aesthetic. A commitment.