What we build
One lab. Three businesses.
Data DAOs, reinforcement-learning infrastructure, and frontier AI products. Three separate businesses, each with its own customers and its own reason to exist. We build all three, and each one stands on its own.
01 / Data DAOs
Data owned by the people who make it
Data DAOs that capture the highest-signal training data, real AI coding sessions and computer-use trajectories, owned and permissioned by the people who create it. Contributors are paid. Models get data they cannot find anywhere else.
Coding dataComputer-useOn-chain provenanceContributor rewards
Data contributionslive
Owned this week1.24 TB
Coding session48 MB · owned+120
Computer-use trace12 MB · owned+40
Terminal workflow7 MB · owned+25
02 / RL infrastructure
Reinforcement-learning gyms
We turn the datasets into environments where a model tries a real task and gets scored on whether it actually succeeded. The infrastructure runs millions of these attempts, so a model learns the job by doing it over and over, not by reading about it.
env · web-nav-v3step 48,120
reward 0.82success 74%rollouts 12.4k
03 / Products
Frontier AI products, starting in finance
We build consumer products on frontier AI, starting with finance. The first is an agent that does your research, makes the call, and places the trade for you. More will follow.
Agent · financegpt-si · 1
Find a low-risk entry on SOL this week
Breakout confirmed above the 50-day. Opened 2× SOL, stop −1.2% / target +3.4%. Executed
What we believe
Models get smart by doing real work. The people who do that work should own the record of it, and get paid when it trains a model.
01
People own their data
Contributors keep ownership of what they record, and earn when it's used. Nothing is taken without consent.
02
Models learn by doing
They practice real tasks and get scored on the result, instead of guessing from scraped text.
03
Products stand on their own
Each product serves real users with real money on the line. It has to win as a product, not as a demo.
04
Three bets, not one
Data, training, and products are separate businesses. Any one of them can succeed on its own.