Common questions
Short answers for everyday Lore usage decisions.
Frequently asked questions
| Question | Answer |
|---|---|
| Is Lore a vector database? | No. Lore uses PostgreSQL, pgvector, full-text search, generated views, and explicit memory metadata to manage agent memory. |
| Do I need Docker? | No. Docker is the easiest path, but you can run the server against your own PostgreSQL and configure clients manually. |
Do I need pg_jieba? | It is recommended for Chinese recall. Lore can fall back to zhparser or simple, but quality may drop. |
| Should every chat result become memory? | No. Store durable facts, preferences, rules, decisions, and workflows that should matter again. |
Should CLAUDE.md and AGENTS.md keep all rules? | No. Keep bootstrap guidance locally and move shared durable rules into Lore. |
| Can several agents share memory? | Yes. That is the main point: one Lore server can serve Codex, Claude Code, Pi, OpenClaw, Hermes, and MCP clients. |
| What if recall is noisy? | Improve memory shape first: split broad notes, clarify disclosure, tune glossary, and lower inflated priority. |