Lesson 411 lessons
Memory — Short-term and Long-term
Short-term memory: the current task context
Within a single agent run, short-term memory is simply the conversation/action history so far — what's been tried, what results came back. This lives only for the duration of that task.
Long-term memory: persisting across runs
Long-term memory stores durable facts or learned preferences in a database, retrieved and injected into future agent runs — allowing the agent to "remember" a user or past decisions across sessions.
Avoiding context bloat
Not everything should be remembered forever. Design a clear policy for what gets stored long-term (key decisions, preferences) versus what's discarded after the task (intermediate reasoning, tool call noise).
Key Takeaways
- Short-term memory is the current task's action history.
- Long-term memory persists durable facts across separate agent runs.
- Store long-term memory in a database, injected into future context.
- Have a clear policy for what's worth remembering vs. discarding.
Design a memory policy
Write down 3 types of facts your agent should remember long-term about a user, and 3 types it should discard after each task.