Specs that survive a procurement review.
Six primitives. Zero magic.
Each module ships with a spec sheet, an SLA, and a rollback path. We built this for engineers, lawyers, and analysts who get audited.
Ingest anything
PDF, DOCX, TXT, Markdown. We parse, chunk, and embed every page at 1536 dimensions.
Cited answers
Every response links back to the exact source document and section. No black-box hallucinations.
23ms retrieval
pgvector + HNSW indexes. Median similarity search over 100k chunks finishes before you blink.
Row-level isolation
Postgres RLS guarantees your workspace data never leaks across tenants. Audited monthly.
Versioned chunks
Re-upload to update without losing chat history. Old citations stay resolvable forever.
Multi-turn memory
Conversations carry context across questions. Refine, drill down, follow threads.
Three steps from upload to answer.
Upload
Drag files into your workspace. We parse, OCR if needed, and chunk at semantic boundaries.
Index
Each chunk gets a 1536-dim embedding. HNSW indexes are built in under 4 seconds per MB.
Ask
Type a question. We retrieve top-K chunks, ground the LLM, and return cited prose in under 2 seconds.
What teams say after week two.
“Replaced four hours of Ctrl-F per week. Citations are what won me over.”
“The retrieval is shockingly good. Our compliance audit took 40% less time.”
“I uploaded twelve years of board minutes. It found a clause we forgot existed.”
“Engineering loved it. Postgres + pgvector under the hood means we can self-host later.”
“The cited-source UI builds trust with our team in a way other tools never did.”
“Onboarded our whole studio in 90 minutes. Now it answers our brand-guideline questions.”