Store and retrieve past agent episodes — full trajectories of thought, action, and outcome — to improve performance over time.
Five passes over the same idea, each from a different angle. Do them in order, or jump to whichever you need.
Episodic memory lets agents learn from their own history. An episode is a complete interaction: the goal, the plan, the tool calls, and the outcome. By storing episodes and retrieving similar ones during planning, agents can avoid past mistakes and replicate past successes. This is the foundation of self-improving agent systems.
