Report an issue

Practice quizzes for Australian citizenship, driving, and immigration topics

Quizzes

By accumulating the history of observations, tool results, and reasoning into its context window (in-context memory)

By writing all intermediate results to a shared database that all agents access

By sending state snapshots to the user after each step for safekeeping

Agents are inherently stateless; each step is an independent LLM call with no memory of prior steps