From agentdb-search
Hybrid search — BM25 keyword + dense vector fused with Reciprocal Rank Fusion. Use when queries have specific identifiers, code symbols, or proper nouns that pure semantic search might miss.
npx claudepluginhub ruvnet/agentdb --plugin agentdb-searchThis skill uses the workspace's default tool permissions.
Combine sparse (BM25, exact terms) and dense (vector, semantic) search via Reciprocal Rank Fusion. Catches both "exact-string" queries and "intent" queries in one call.
Fuses vector and keyword search results using RRF, linear combination, and reranking for better recall in RAG systems and search engines.
Combines vector similarity and keyword search for improved retrieval recall. Use when building RAG systems, search engines, or handling specific terms pure vector search misses.
Combines vector similarity and keyword search for improved retrieval in RAG systems, search engines, or when single methods lack recall.
Share bugs, ideas, or general feedback.
Combine sparse (BM25, exact terms) and dense (vector, semantic) search via Reciprocal Rank Fusion. Catches both "exact-string" queries and "intent" queries in one call.
getUserById, JWT_SECRET)agentdb_hybrid_search(
query: <string>
k: 5
weights?: { bm25: 0.5, vector: 0.5 } // default; bandit can pick
filters?: { ... } // metadata filters
)
Each result gets a rank in BM25 list (r_bm25) and a rank in vector list (r_vec). Final score:
score(i) = w_bm25 / (60 + r_bm25(i)) + w_vec / (60 + r_vec(i))
The constant 60 (k in the RRF paper) prevents top-1 hits from dominating; the additive structure rewards being well-ranked in both lists.
weights manually per query. Let the bandit (weights: 'auto') learn the right balance.