The first artifact a research lab ships says something about how the lab thinks about its work. Ours is a harness — a piece of operator software that runs on your machine and stays out of your way. It is not a chatbot. It is not an orchestrator. It is the thing you sit in front of for hours when you are trying to figure something out.
We started here because the work we want to do downstream — training models that operate against live, reflexive, on-chain markets — is impossible without a substrate that captures how operators actually work. Trades, sessions, watches, runbooks, the trail of why a thesis was opened and what closed it. A research lab without that substrate is reading screenshots.
So the harness comes first. Operators use it. The work credits a token rail that pays operators back for the sessions they run. The substrate compounds — both as a product surface and as a corpus of operator behavior the lab can study, label, and learn from with consent.
A few principles that shape what we work on:
- Operators own the work. Local-first, structured ledger on the operator's machine, no remote unless opted in. Sharing is attributed.
- Read by default. The harness is read-only out of the box. Anything that mutates an account is gated behind explicit approval.
- Hyperliquid and Polymarket are peers. One cockpit holds both. A thesis can span perps and prediction markets in one session.
- Tools before models. We will say more about the second half later. For now, the harness is the work.
If you are running operator workflows against on-chain markets and you want to hold context across sessions instead of losing it every time you close a tab — that is what the harness is for. Read the docs for specifics. The roadmap lives there too.
— Axe Labs