What is meant by agent memory?

• A. A temporary cache that is cleared after each task.
• B. A logging file for audit trails only.
• C. A storage layer that preserves past interactions, tool outputs, and context to maintain continuity.
• D. A fallback mechanism to retry failed tool calls.

Answer: (C)

Memory in an agent is a storage layer that preserves past interactions, tool outputs, and context to maintain continuity. By keeping a record of what happened earlier, what tools produced, and what the user is aiming for, the agent can refer back, stay coherent across multiple steps, and tailor actions to the user’s goals. This enables multi-step tasks to flow smoothly without re-asking for information or losing important details.

Some descriptions describe a temporary cache that resets after each task, or a log file kept for audits, or a mechanism to retry failed tool calls. Those serve different purposes—history, records, or robustness—but they don’t capture the ongoing preservation of context and past results that memory provides for maintaining continuity throughout the interaction.