Hybrid Retrieval
- Dense vectors plus BM25
- Filters applied inside the index
- Reciprocal rank fusion built in
- One round trip, one response
Similarity, keywords and metadata resolved together in a single query. No fan-out to three services, no re-ranking layer of your own, no post-filter that quietly ruins recall.
One query, not three services
Most teams bolt a keyword index next to a vector store and merge the results in application code. Nodeform runs both inside one engine and fuses them with reciprocal rank fusion, so filters narrow the search rather than throwing away its answers.
Why hybrid retrieval matters
Filters that do not cost recall
Metadata predicates are evaluated during graph traversal, not after it. A one-percent-selective filter still returns a full result set.
Exact terms still work
Part numbers, error codes and names are the queries embeddings are worst at. BM25 catches them in the same call.
One service to operate
A keyword cluster you no longer run, a merge layer you no longer debug, and one latency number instead of three.
Ready to stop running
your own vector index?
Read the quickstart, or talk to the engineer who will run your index.
