ASE
Frontier inference, held to one contract.
A language-model provider. Frontier inference behind one endpoint, metered and priced like infrastructure rather than like a product.
We form partnerships and hold them past the point at which most parties stop holding.
ScrollCompanies we build inside and hold. Each one is a position we operate, not an entry in a portfolio.
Eight heldFrontier inference, held to one contract.
A language-model provider. Frontier inference behind one endpoint, metered and priced like infrastructure rather than like a product.
The feed, made by the people in it.
Creative social media, where the making is the point. Every post carries the tools that produced it, so the work can be taken apart, rebuilt and posted back.
A company that runs itself, and shows its working.
Agentic business management. Agents hold the routine load — scheduling, reconciliation, reporting, the third follow-up nobody sends — and every action is logged, reversible and attributable.
Read the chain before you take the position.
DeFi research, and the instruments that come out of it. Pool depth, counterparty exposure and the real cost of exit, measured continuously and published as numbers rather than opinions.
Where the spread actually goes.
Currency exchange research. We take a rate apart — spread, routing, settlement lag, the fees that never reach the quote — and publish what it costs to cross a border.
A chain small enough to keep in your head.
A fast, lightweight, scalable blockchain. Sub-second finality on hardware you already own, a node small enough to read end to end, and a fee model that does not punish being early.
Agents that find each other and settle themselves.
An agentic trading network. Autonomous counterparties find one another, negotiate terms and settle with no venue in the middle — every agreement written where both sides can see it.
Findings first. Product later, if at all.
Deep research into how language models reason, remember and fail. Weights, evaluations and negative results are published in the open rather than held back as product.
We assemble the party before we assemble the deal. Most structures fail because they were drawn around incentives that had not yet been said out loud. We say them first, and we write them down where both sides can see them.
A partnership is an obligation that has to be maintained. We hold the maintenance work — reporting, arbitration, the slow correction of drift — for as long as the relationship remains worth holding.
Software is where an agreement becomes observable. Our engineering practice builds the instruments a partnership needs in order to see itself: ledgers, terminals, controls, and the reporting that sits under all three.
Every arrangement contains more capacity than its parties use. We locate it, price it, and route it back into the position rather than out of it.
We read the counterparty before we read the opportunity. Nothing is proposed in the first six weeks.
Terms are drafted against stated incentives, not against precedent. Where the two disagree, precedent loses.
The agreement is given a surface. Positions, obligations, and thresholds become things both parties can look at on the same screen.
We run inside the partnership rather than alongside it. The work is ordinary and continuous.
Every arrangement is repriced on a fixed cadence. Continuation is a decision, taken deliberately, and it is taken often.
Value does not accrue to the party with the best terms. It accrues to the party still present when the terms stop mattering.
Most arrangements are priced at formation and never repriced. We take the opposite view: a partnership is an instrument whose value is discovered slowly, in the ordinary work of keeping it intact. That work is the practice.