Vector retrieval, no generation step
Search by meaning,
not keywords.
Zero-LLM Search embeds your documents into vector space and retrieves by intent. No model rewrites, summarizes, or hallucinates on top of your results — what you get back is what's actually in your data.
Process
Three steps, no generation layer anywhere in between.
01
Upload
Drop in a JSON file of documents — each one becomes a searchable record in your knowledge base.
02
Embed
Every document is converted into a 384-dimension vector with BGE-small-en-v1.5 and indexed in Qdrant.
03
Retrieve
A query is embedded the same way, then matched by cosine similarity — nothing is generated or reworded.
Why it's different
What "no LLM" actually buys you.
Grounded
Every result traces back to a real document in your dataset — there's no generation step that can drift from the source.
Fast
A similarity lookup against an indexed vector store returns in milliseconds, without waiting on a model to compose a response.
Inspectable
Every hit ships with its similarity score, so you can see exactly how confident the match is rather than trusting a generated summary.