Niches · Retrieval & vector search
Find by exact terms and meaning together
Four or more companies' docs show a full fit.
What counts as done
One query combines keyword and vector scores, an exact identifier match ranks first when present, and the keyword-against-vector weight can be set per query.
Measured, for our own tests: on a mixed query set hybrid beats vector-only recall at 10. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: find the clause that mentions section 4.2
, search for SKU AB-1234 and similar products
, total assets FY2024 in the 10-K
, typo-tolerant search over the catalogue
Does the job 5
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Elasticsearch C Elasticsearch B.V. | The rrf retriever fuses lexical and kNN child retrievers in one request, each with an optional weight, and the linear retriever normalises and sums scores; both are set per query. source |
| Meilisearch A Meili SAS | Each search takes a hybrid object with an embedder and semanticRatio from 0 to 1, merging keyword and vector ranking in one query, and the built-in MCP searchInIndexes accepts hybrid, which #18 lacked in the older server. source |
| Milvus and Zilliz Cloud API + MCP B Zilliz Inc. | hybrid_search combines a built-in BM25 sparse field with dense search in one request, merged by RRFRanker or WeightedRanker with weights set in the request. source |
| Qdrant API + MCP BB Qdrant Solutions GmbH | One /points/query call fuses a BM25 prefetch and a dense prefetch with rrf or dbsf, and RRF takes per-ranking weights in the query itself. source |
| Weaviate API + MCP C Weaviate B.V. | Hybrid fuses BM25F and vector results in one query with alpha set per query (0 keyword to 1 vector) and a choice of fusion method; the MCP tool exposes alpha too. source |
Does the job, with a limit 9
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Chroma API + MCP D Chroma Inc. | The Search API fuses rankings with RRF and per-ranking weights in one request, but the docs say it is available in Chroma Cloud only, not on single-node or local Chroma. source |
| Cloudflare AI Search BB Cloudflare, Inc. | Instances index for vector and keyword search by default and merge them with fusion_method rrf or max, but the fusion is set in instance configuration, with no per-query keyword-against-vector weight. source |
| Epsilla Vector Database E Epsilla, Inc. | Hybrid search is assembled client-side with add_retriever for dense and sparse retrievers and an RRF reranker, not one server-side query, and no keyword-against-vector weight is shown. source |
| LanceDB B LanceDB Systems, Inc. | query_type=hybrid fuses vector and full-text results with RRF by default, but only after an FTS index is built, the use case's named blocker, and no per-query weight is shown on the hybrid page. source |
| OpenSearch BB OpenSearch Software Foundation, a project of The Linux Foundation | A hybrid query combines keyword and neural clauses, merged by a normalisation processor or RRF score ranker, but the weights array lives in a search pipeline set up beforehand, not in the query itself. source |
| Pinecone API + MCP A Pinecone Systems Inc. | One request can filter by exact terms then rank by meaning, but a vector clause must be the only scoring clause, so fusing BM25 with vector scores is separate searches merged client-side with RRF; no per-query weight in one call. source |
| Typesense API + MCP BB Typesense, Inc. | Putting the embedding field in query_by beside text fields fuses keyword and vector ranks, with vector_query alpha per query, but the silent 128-token truncation (#3004) is this use case's named blocker, and #2618 shows short hybrid pages after deletions. source |
| Upstash Vector API + MCP B Upstash, Inc. | A hybrid index queries its dense and sparse parts in one call and fuses them, with fusionAlgorithm (RRF or DBSF) and queryMode per query, but there is no weight to set keyword against vector. source |
| Vespa BB Vespa.ai AS | Rank profiles combine bm25 and vector closeness, with a global-phase reciprocal_rank fusion, but the balance is written in the deployed rank profile, and #33421 asks for per-query thresholds without redeploying the schema. source |
What stops agents finishing
- hybrid exists in the REST API but not in the MCP tools
- keyword side missing until a separate full-text index is built
- built-in embedder truncates input without warning
Around it
Before: vector-search.ingest-sync. After: vector-search.rag-over-docs.
Where this comes from
Weight 16 of 100 in Retrieval & vector search: 10 of 37 accounts (Weaviate alpha, Meilisearch and Typesense comparisons, OpenSearch and Elasticsearch hybrid use, LanceDB and Vespa tuning)
The same as data: /api/v1/use-cases/vector-search.hybrid-keyword.json, or the MCP tool get_use_case with id vector-search.hybrid-keyword. Other jobs in this category: /api/v1/niches?category=vector-search.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.