# Agentic memory

Question: How should an AI agent remember what it learned in past conversations, so it does not start from zero every time?

Status: ongoing. Started 2026-03-18. Last entry 2026-03-18. 1 entries, 13 min read, 4.7k tokens.

Agents that forget everything between conversations repeat the same investigations. This journey is about giving them a memory that stays useful as it grows.

## What we think now

Revision 1, 2026-03-18.

1. Plain vector search over old conversations is not enough. It has no relationships, no way to retire outdated knowledge, and no ranking by what has proven useful.
2. Store memories as a graph, with 5 memory types and 7 typed edges (prerequisite, enables, alternative, extends, supersedes, conflicts, related).
3. Curate after every conversation. Extract 0 to 5 patterns, find similar memories with keyword and semantic search merged by reciprocal rank fusion, then store, combine or discard.
4. Trust comes from use. Four usage signals boost a memory's rank by 5 to 50%. One repeated diagnosis went from 12 minutes to 8 seconds.

## Open questions

- Over months, does the graph get sharper or does it drift?
- What should happen to a memory that keeps being contradicted?

## Entries

1. [Agent Memory Meets Graph: Introducing MemGraph — Long-Term Memory for AI Cloud Agents](https://engineering.cloudthinker.io/journeys/agentic-memory/01.md) (2026-03-18, start, 13 min, 4.7k tokens)

Web version: https://engineering.cloudthinker.io/journeys/agentic-memory/