Memory Search (SOTA)

Strategy Selection

• Default (dense): fast single-hop semantic match
• --hybrid: sparse + dense with RRF fusion (20-49% better for keyword+semantic queries)
• --graph-rag: multi-hop knowledge retrieval (30-60% better for reasoning queries)

Steps

1. Parse query and flags — extract search text and strategy flags from arguments
2. Select retrieval strategy:
   Dense search (default):
   bash

   npx @claude-flow/cli@latest memory search --query "QUERY" --namespace NAMESPACE --limit 10

   Or via MCP:

   mcp__claude-flow__memory_search({ query: "QUERY", namespace: "NAMESPACE", limit: 10 })

   Hybrid search (when --hybrid or query has specific keywords):
   bash

   npx ruvector search "QUERY" --hybrid --limit 10

   Graph RAG (when --graph-rag or multi-hop reasoning needed):
   bash

   npx ruvector search "QUERY" --graph-rag --limit 10

   Smart retrieval (when --smart or complex recall needed):
   bash

   npx @claude-flow/cli@latest memory search --query "QUERY" --smart --limit 10

   Or via MCP:

   mcp__claude-flow__memory_search({ query: "QUERY", smart: true, limit: 10 })

   Applies 5-phase pipeline: query expansion, RRF fusion, recency boost, MMR diversity, session round-robin. Best for: multi-session recall, temporal queries, diverse result sets.
   Unified cross-namespace:

   mcp__claude-flow__memory_search_unified({ query: "QUERY", limit: 10 })

3. Apply MMR reranking — for diverse results, filter near-duplicates (cosine > 0.92) while maximizing relevance
4. Apply recency weighting — boost recent entries with exponential decay (0.95/day)
5. Synthesize context (for complex queries):

   mcp__claude-flow__agentdb_context-synthesize({ query: "QUERY", sources: ["patterns", "tasks", "solutions"] })

6. Present results — ranked by composite score (relevance * diversity * recency), with source namespace attribution

Namespace Guide

patterns

tasks

solutions

feedback

security