TYPES OF AGENT MEMORY

SHORT-TERM MEMORY INSIDE THE CONTEXT WINDOW Short-term memory in an agent is whatever is currently inside the context window, the limited amount of text the model can read in one go, which you studied in Part Five. It includes the system prompt, the recent messages of the current conversation, the results of recent tool calls, and any documents placed there. The model can use all of this directly and instantly. But short-term memory has two limits. First, it is size-limited. When the window fills, older material must be dropped or compressed, as you learned in Chapter 21 on compaction and summarization. Second, it is temporary. When the run ends, the context window disappears, and everything in it is gone unless the harness saved it somewhere. Short-term memory is like a whiteboard in a meeting room. It is great for the current discussion, but someone will wipe it clean afterward.