States

The runtime is also responsible for maintaining each agent’s long-term operational state. This includes tracking allocations, open positions, protocol exposures, execution history, pending actions, and any other data required for consistent autonomous behavior. By storing this state outside the agent’s reasoning loop, the runtime ensures that every decision is based on accurate and up-to-date context.

As agents act, the runtime updates their state to reflect new conditions or outcomes. This persistent memory allows agents to behave as long-lived systems rather than isolated tasks. They can adapt their behavior over hours, days, and weeks, and maintain continuity even as market conditions fluctuate. State management also provides transparency to users, since the runtime can surface performance and activity histories.