Organizations that compile.

We build tools for autonomous organizations
and benchmarks to see what AI can really do.

An organization needs a purpose, roles, and rules for making decisions. We want those things to be clear enough to describe in code and run as a working system.

02 / Products

Meet the tools
001

vsmlite

Small steps toward autonomy.

One parent organization helps a new one learn to work on its own. Responsibility grows with evidence.

002

proofstate

Let the work speak.

An open benchmark platform for independent researchers. Compare AI systems through tasks, traces, and public results.

VSM HARNESS INDEX

Track autonomy,
function by function.

Evidence-backed fingerprints with dated reviews and pinned GitHub refs.

HarnessS1S2S3S3*S4S5
HeadcountAAACAA
OmnigentAAAA
LoopXAAA(P)AP
OpenHarness (HKUDS)AAAA

04 / Research

Beyond familiar patterns

ARCTIC

Small models. New reasoning tasks.

Explore visual puzzles that test rule discovery, state tracking, and planning
Then compare how small models solve them.

85 reasoning tasks / 4 difficulty levels