Skip to content
LACE
  • v0.1 Current
  • Python
  • TypeScript Soon

Platform Guide

Search & RAG

Hybrid retrieval fuses lexical and vector legs with reciprocal-rank fusion, filters by source ACLs, and optionally reranks on GPU.

Hybrid retrieval

A lexical leg and a vector leg run against the same corpus and their rankings are combined with reciprocal-rank fusion. Lexical catches the exact identifier, part number, or clause reference that embeddings blur; vector catches the paraphrase that keyword search misses. Fusing them avoids having to choose which failure mode to live with.

Permissions-aware by construction

Results are filtered by source ACLs during retrieval, not after generation. A document the requesting identity cannot access is never a candidate, so it cannot be summarized, quoted, or leaked through a citation. The same holds for agents acting on a user's behalf.

Evidence identity

Evidence carries stable document, block, and span identity. Answers open the original passage rather than a reconstruction of it, and tables stay tables — a figure is returned from its actual cell with its row, column, and units intact instead of being flattened into prose and re-read by a model.

See Enterprise Search for the product surface built on this.