We built three memory systems. The most expressive one wasn’t the most useful.

Efforts to improve Eon, Tiger Data’s Slack agent, led to three memory-system designs aimed at retrieving relevant context without overwhelming agents with irrelevant information. An initial user-scoped list of text facts was simple but limited in sharing, organization, and proactive capture, while a later PostgreSQL knowledge graph offered rich metadata, temporal data, and relationships but required agents to define inconsistent edge semantics and perform complex SQL operations. A third design replaced the graph with an ltree-based hierarchy in PostgreSQL, making storage, navigation, sharing, and hybrid BM25-plus-embedding search more straightforward while allowing agents to choose what to save and retrieve without managing database mechanics. Although the tree could not express every relationship, targeted metadata links were added only for specific needs such as recovering surrounding conversation turns or Git history. As teams adopted the system for RAG use cases including data catalogs, marketing resources, documentation, and coding context, differing user and permission models showed that access policy should remain application-specific. These lessons led to searchgres.js, an open-source TypeScript library that extracts reusable PostgreSQL search, embedding, indexing, and filtering infrastructure while leaving applications responsible for information organization, permissions, workflows, and use of retrieved results.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.