Platform / Vector Search & Retrieval

Retrieval with citations, in one declaration.

Every Maestra app includes a managed retrieval pipeline: extraction, chunking, embedding, hybrid search, and reranking behind a single declaration, returning answers with the passages they came from. Builders point it at documents. The platform does the rest.

01

The whole pipeline, managed

Retrieval pipeline
Sources
PDFs
Docs
Tickets
extract
Index
01Chunk
02Embed
03Store
hybrid search
Answer
Reranked passages
Citations
Every stage is handled by the platform and re-runs when sources change.
02

One declaration

export const Policies = collection("policies", { sources: [drive("/HR/Policies")], search: "hybrid", rerank: true, }); const hits = await Policies.ask("How much parental leave?");
03

Hybrid search, reranked

Keyword and semantic search run together, so exact terms and loose phrasing both find the right passage. A reranker then orders results by how well they actually answer the question.

Results · “How much parental leave?”
Leave policy §4.20.94
Benefits guide p.120.71
Onboarding FAQ0.52
Holiday calendar0.18
Scores after reranking. The top passage is returned with its citation.
04

The guarantees

Cited
Every answer carries the passages it came from.
Scoped
Search respects the caller’s role; people only retrieve what they can read.
Fresh
Indexes update automatically when sources change.

Start building on Maestra.

Everything you just read is standard in every app, running from the first deploy.

Start building →
← Back to the Platform