Lore
Documentation navigation
Docs/Use Lore/Improve recall quality

Improve recall quality

Tune glossary terms, disclosure triggers, priority, generated views, and memory shape.

Improve recall by improving memory shape before changing thresholds.

Improvement order

Use this order:

  1. Make the memory focused: one topic per node.
  2. Give it a specific path.
  3. Write a concrete disclosure trigger.
  4. Add glossary terms users and agents actually type.
  5. Set priority honestly.
  6. Check that embeddings and View LLM settings are configured.
  7. Tune recall settings only after the content is clean.

Common problems:

SymptomFirst fix
useful memory never appearsadd disclosure and glossary terms
too many broad resultssplit large memories and lower priority
Chinese queries are weakverify pg_jieba or zhparser; otherwise Lore falls back to simple tokenizer
semantic recall is missingconfigure embedding provider and rebuild generated views
results look stalerun Dream or refresh generated memory views after major edits
Web ConsoleLearn what each main page in the Lore Web Console is used for.Recommended routineFollow a simple routine for memory review, recall checks, Dream, and backups.Write a durable memorySave one useful memory with path, priority, disclosure, glossary, and content.Use recall in chatRead the recall block as memory pointers and open useful memories before relying on them.