Agent memory
Agent memory: Information retained across steps or sessions to preserve state, preferences and useful results for an agent. Working memory covers the current task; long-term memory requires selection, updating and deletion. Keeping everything in context increases cost and the risk of stale data.
In italiano: Agent MemoryLegge III · Come una rete impara dai datiWorking memory covers the current task; long-term memory requires selection, updating and deletion. Keeping everything in context increases cost and the risk of stale data.
Examples
- Conversation state
- Revocable user preferences
- Results of previous actions