AGT-04 · Agent memory: design, set up, retain and erase¶
Teams mix chat history, user facts and past episodes in one store, so memory grows without limit and cannot be cleanly erased. Privacy requests and schema changes then turn into risky manual database work.
- Who reads it: GenAI engineers designing memory and DevOps engineers and data protection staff who run and audit it
- Outcome: Memory with clear retention and erasure done in hours, not weeks
- Pack owner: James (catalog owner role: Agent Lead)
Master catalog entry¶
- Catalog name: Agent Memory Framework (P2)
- Why it matters: Continuity across sessions is what users perceive as intelligence
- What it removes: Short-term, long-term and episodic memory decided once
- Stack: Postgres, Redis, Azure AI Search, LangGraph store
- Maturity target: Level 2 — includes retention, summarisation and privacy handling
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-AGT-04-01 | AGT-04-S1 | Choose memory layers and stores: short-term, long-term and episodic | Decision guide | Medium | 3 | draft | James | 2026-12-08 |
| PB-AGT-04-02 | AGT-04-S2 | Set up LangGraph checkpointer and store on Azure Database for PostgreSQL | How-to | Medium | 3 | draft | Sanjay Sarvesh | 2026-12-24 |
| PB-AGT-04-03 | AGT-04-S3 | Erase a user's agent memory on a data subject request | Runbook | Medium | 4 | draft | Sanjay Sarvesh | 2026-12-11 |
| PB-AGT-04-04 | AGT-04-S4 | Run memory retention, growth control and state schema migration | Runbook | High | 4 | draft | James | 2026-12-03 |
| PB-AGT-04-05 | AGT-04-S5 | Memory privacy and retention review checklist | Checklist | Low | 3 | draft | Sanjay Sarvesh | 2026-12-22 |