A stateful agent has a durable identity and an associated record of relevant task state. New messages and tool results are applied to that state, allowing the agent to continue a session rather than rebuilding all context from scratch on every request.
State needs clear ownership and ordering. Isolated histories and serialized updates help prevent concurrent events from overwriting each other or producing a session state that never existed in a valid sequence.
ELI5
A stateful agent remembers useful information between steps or sessions instead of starting from nothing each time. Its saved state can include progress, earlier decisions, preferences, and unfinished work.
For example, a weekly reporting agent can remember which sources were processed last time and continue from the newest verified point. It must also recognize stale or user-specific state so old information does not cause a wrong action.
Where is a stateful agent's state stored?
It is usually stored in durable infrastructure outside the temporary model call, such as a database or event-backed state object.
Why isolate each stateful agent session?
Isolation prevents unrelated sessions from mixing data and lets each history be updated and recovered independently.
