Committed
Trading agent runtime
The execution loop an AI agent needs to trade on someone’s behalf, rather than the loop a human interface assumes. A person can watch a screen and react. An agent cannot — it can only act on what the venue is able to prove. That gap is not closed by an API; it is closed by a runtime with defined semantics at every step: predict before committing, submit without ambiguity, resume after interruption, and read back a record that settles what happened.
The account-model work sits alongside this: authorization today is all-or-nothing with an expiry, and richer terms — what an agent may hold, how much it may lose, what it may do next — are what turn a permission into a mandate. See AI trading.
Institutional custody integration
Direct funding from institutional custody, starting with Fireblocks. The blocker for institutional capital is rarely the venue’s mechanics. It is that the capital cannot leave its custody arrangement to reach the venue. Integrating at the custody layer means an institution funds and trades without moving assets out of the controls it is required to operate under.Market surface
Liquidity and capital
Access and accounts
Protocol
Data and observability
Nothing here carries a date. Dates belong to Milestones, which covers the two commitments that do have them. Items move from this page to the changelog when they ship.
Where to go next
Milestones
What is running today, and the two dates that matter.
Protocol changelog
Dated record of what has shipped.
AI trading
Why the agent runtime is the shape it is.
Architecture
What these capabilities are being built on.