Skip to main content

​
Hybrid Search

Memorose retrieval combines multiple signals rather than relying on one search path.

​
What Happens During Retrieval

  1. The query is embedded
  2. Vector retrieval runs against stored memory units
  3. Text retrieval runs against Tantivy
  4. Optional graph expansion uses graph_depth
  5. Shared organization knowledge can be included
  6. Results are merged into one ranked response

​
API Example

curl -s -X POST "http://127.0.0.1:3000/v1/users/user_123/streams/$STREAM_ID/retrieve" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "org_id": "default",
    "agent_id": "coding-assistant",
    "query": "How should I help this user with Rust async work?",
    "limit": 10,
    "enable_arbitration": false,
    "min_score": 0.2,
    "graph_depth": 2
  }'

​
Useful Filters

  • org_id: include organization-scoped shared knowledge
  • agent_id: bias toward one agent’s procedural memory
  • start_time and end_time: valid-time filtering
  • as_of: point-in-time transaction filtering
  • graph_depth: expand from directly matched memory into related graph context
enable_arbitration is a request-level retrieval option. It does not choose the server-side reranker mode. Configure weighted, arbitrator, or HTTP reranking in [reranker] or through MEMOROSE__RERANKER__* environment variables.
  • Preference and profile recall
  • Mixed factual and procedural lookup
  • Queries that need both exact terms and semantic similarity
  • Retrieval that benefits from graph neighbors or organization knowledge