Learning path · Agents & Orchestration · 67
Agent Memory Tiers
Layered memory—working buffer, episodic summaries, long-term vector store—for agents across sessions and tasks.
Why it matters
- Prevents agents from forgetting constraints mid-run.
- Separates ephemeral tool output from durable user facts.
- Must respect privacy retention and deletion requests.
Key ideas
- Working memory
- Episodic summaries
- Long-term retrieval
Top resources
- 01DocsLangChain
LangGraph
Why this resource. Working state vs persisted checkpoints vs retrieved long-term memory.
Covers in this concept
- checkpoint
- store
- 02ArticleAnthropic
Building effective agents
Why this resource. Which memories should be files and search, not more tokens in the prompt.
Covers in this concept
- memory tiers
Tier agent memory like human workflows: scratchpad for current plan, session summary for recent context, vector store for user preferences and past cases. Write policies for promotion—what may move from working to long-term memory. Encrypt and tenant-isolate stores. On user delete requests, purge all tiers consistently. Test GDPR deletion end-to-end across tiers quarterly; episodic summaries often linger after users expect full erasure. Run quarterly deletion drills across working, episodic, and long-term stores together.
Updated 2026-08-09 · Full learning path