Agent Memory allows an agent to retain information beyond its current execution and use it in future interactions.
For example, an agent might remember:
- User preferences
- Previous decisions
- Past interactions
- Learned facts
- Previous experiences or outcomes
A useful distinction is:
State represents what is happening now; memory represents what can be retained for later.
Memory can be stored in different forms, such as:
- Conversation history
- Databases
- Vector stores
- Structured user profiles
- Experience or event logs
The important part is that memory does not mean putting everything into the context. Relevant memories should be retrieved when needed and added to the current context.
Memory stores information; retrieval decides what becomes context.