NewMCP ServerView docs
Hybrid Search

Find everything.
Miss nothing.

Our three-way retrieval system combines semantic understanding, contextual awareness, and keyword precision via Reciprocal Rank Fusion for comprehensive results.

3x
Retrieval Methods
Sub-sec
Target Response
100+
Languages Supported
RRF
Rank Fusion
How It Works

Three retrieval methods, one unified result

1

Dense Vector Search

BGE-M3 embeddings capture semantic meaning, finding conceptually similar content even without keyword overlap.

2

Sparse Vector Search

Contextual term weighting identifies important phrases and their relationships for precise matching.

3

BM25 Keyword Search

Classic term frequency analysis ensures exact matches and specific terminology are never missed.

Reciprocal Rank Fusion

How three methods combine

Dense Vector
94%
Sparse Vector
89%
BM25 Keyword
87%
RRF Fusion
3x
Retrieval Methods Combined
Developer Experience

Simple API, powerful results

One API call gives you access to our full hybrid retrieval pipeline. Configure retrieval methods, enable reranking, and get cited results.

search.ts
// Search with hybrid retrieval
const results = await lakehouse.search({
  query: "Q4 revenue drivers",
  options: {
    retrieval: "hybrid",  // dense + sparse + bm25
    rerank: true,         // cross-encoder reranking
    limit: 10,
  }
});

// Results include relevance scores and citations
results.forEach(doc => {
  console.log(doc.title, doc.score, doc.citations);
});

Experience hybrid retrieval

Start your free trial and see the difference hybrid search makes.