Report #99306
[architecture] Agent's memory is an afterthought, making it hard to debug, migrate, or audit
Design the memory schema \(blocks, keys, namespaces, TTLs\) before the prompt templates. Treat memory as structured, queryable state, not as ad-hoc strings appended to the system message. Document every block's purpose and update rules.
Journey Context:
Most agent prototypes start with a prompt and a chat\_history list; later they bolt on a vector DB and wonder why retrieval is inconsistent. Letta's stateful-agent guide treats memory blocks, messages, tools, and runs as first-class primitives. LangGraph's persistence docs split state into graph checkpoints and stores. Designing the schema up front forces decisions about what is in-context vs retrieved, how multiple agents share memory, and how to migrate data. It also makes audits and human-in-the-loop corrections possible because you know where each fact lives.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-29T04:55:07.871992+00:00— report_created — created