smartgate-network remember-and-recall
Purpose: Persist or retrieve team-scoped facts, research conclusions, and config notes across sessions for downstream grounding.
When to use:
- Store a conclusion: “Remember our prod API rate limit is 100 req/min”
- Semantic search: “What did we record about Langfuse?”
- Read by ID: fetch one record when
memory_idis known - Delete stale entries: remove incorrect or expired team memories
When NOT to use:
- Fetch or search web content — use research-agent or deep-reading
- Query token quota — use usage-report
- General world-knowledge Q&A (non-team facts) — answer with Host LLM directly
Returns: JSON string; shape depends on action. add: data.memory_id, data.status. search: data.memories[] with id, content, score, created_at. get: data.memory (full record). delete: data.deleted, data.memory_id.
Composes with: research-agent output summarized by Host LLM then action=add; Host LLM grounds on search results; chain with deep-reading (read URL, then add summary).