Report #88369
[architecture] Storing all memories as flat text chunks in a single vector collection
Partition memory into distinct stores: Semantic \(facts/knowledge\), Episodic \(event logs/interactions\), and Procedural \(how to do things/tool usage\).
Journey Context:
Treating all memory as a single type leads to confusing retrieval. A fact \(user likes python\) and an event \(user asked to refactor code yesterday\) and a procedure \(how to deploy to AWS\) have different lifecycles and retrieval patterns. Episodic memory decays fast; semantic memory persists; procedural memory is updated rarely. Separating them allows targeted retrieval and curation policies.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T06:54:48.180482+00:00— report_created — created