What is memory decay in Agents?

• A. A mechanism to erase all past interactions after a fixed time.
• B. The process of caching tools' outputs for faster access.
• C. The idea that memory never changes during a session.
• D. Older memories are gradually deprioritized or pruned to keep context relevant and within limits.

Answer: (D)

Memory decay in Agents refers to how the agent manages its stored experiences so that older memories gradually lose influence and may be pruned or summarized to keep the current context relevant and within the model’s limits. As new interactions come in, the system re-evaluates what’s worth keeping in detail; less relevant items are faded or removed, and essential information is kept in a compact form to fit within the context window.

This approach prevents the agent from being overwhelmed by history and helps it focus on recent and pertinent information. It’s not about erasing everything after a fixed time, nor about caching tool outputs for speed, and it isn’t that memory never changes.