Agent memory is not one single feature

sounds like one feature, but it is really a set of different methods that do different jobs. Most real agents do not need every kind of memory; they usually use only two or three that fit the task.

One important kind is , which means the information currently inside the model’s . This matters in long chats, , or multi-step work where the agent must keep track of the current goal, recent changes, and next steps.

The project Letta treats the like RAM: it keeps the most important details close by and moves older information into searchable storage. The tradeoff is that the has limits, and more tokens can raise cost.

Key points

  • is several methods, not one magic feature.
  • Most agents only need a few memory methods for a given .
  • is the information the model can see right now.
  • Long tasks work better when the current goal, recent actions, and next step stay visible.
  • A is limited, and more tokens can increase cost.
Read original