Your documents,
turned into an expert
you can prove.
Ask anything across your contracts, policies and regulations in plain language. Get a clear verdict, the exact source and page, and the precise cost — grounded only in your own documents, never invented.
Q. Can we terminate the vendor agreement early without penalty?
Early termination for convenience is permitted, but the agreement requires 90 days' written notice and settlement of fees accrued to the termination date. Termination for cause carries no penalty if the breach is uncured after the notice window.
The answers your teams need are buried in thousands of PDFs nobody has time to read.
The guesswork tax
People act on half-remembered rules, ask the wrong person, or wait days for someone who's read the document.
The stale-law risk
A regulation was superseded two years ago — but it's still the version being quoted in a decision today.
The AI trust gap
Generic chatbots sound confident and cite nothing. In a regulated business, an unprovable answer is worthless.
Three guarantees built into every answer.
Not features bolted on — the operating principles the whole system is designed around.
It can prove every word.
Answers lead with a verdict — Yes, No, It depends, or “Not addressed in the documents” — then back it with page-level citations. Click one, the original opens at that exact page. When the source is silent, it says so.
Your data never leaves.
Single-tenant, on infrastructure you control. Embeddings run locally on your own hardware. Nothing is used for training, and no third party learns from your contracts. The version security and legal actually approve.
It avoids the expensive AI on purpose.
A four-tier engine (semantic cache → cheap open-weight → mid-tier → premium) answers most questions without ever calling a frontier model. Every answer shows its route and exact cost. Predictable unit economics, at scale.
Built for teams where an unprovable answer is worthless.
Legal & General Counsel
Contract Q&A, MSA reviews, obligation extraction, risk-flagging across the portfolio.
Compliance & Risk
Policy Q&A, regulatory change monitoring, control-mapping to standards (SOC 2, ISO, HIPAA).
Banking & Insurance
Underwriting file review, claims decisioning, loan-document extraction, KYC file summarisation.
Healthcare & Pharma
SOP Q&A, clinical protocol lookup, medical-records extraction, HIPAA-safe deployments.
What ships in the box.
Live inside a regulated enterprise.
Name under NDA.
Deployed in production for a top-tier regulated enterprise — reviewing thousands of contracts, policies and regulatory documents daily. Single-tenant on their private cloud, air-gapped from public internet. Full reference available under NDA on request.
Frequently asked.
What is Document Intelligence?+
Document Intelligence is an Origami-built enterprise document intelligence platform. It turns your contracts, policies and regulations into a Q&A expert that gives verdicts (Yes / No / It depends / Not addressed) with page-level citations back to the original documents. Sovereign, single-tenant deployment; embeddings run on your infrastructure.
How is Document Intelligence different from ChatGPT or generic RAG?+
Three ways: (1) Grounded — it will say "Not addressed in the documents" instead of hallucinating. (2) Sovereign — single-tenant, on-prem or private cloud, data never used for training. (3) Cost-transparent — a four-tier engine (semantic cache → cheap → mid → premium) routes most questions cheaply and shows the route + exact cost on every answer.
Where does Document Intelligence deploy?+
Single-tenant on infrastructure you control — your VPC on AWS / Azure / GCP, or on-prem. Embeddings and the retrieval index stay in your environment. Optional air-gapped mode for regulated deployments.
Which document types does it support?+
PDFs, Word documents, scanned images (OCR), structured spreadsheets, contracts, MSAs, policies, SOPs, regulations, clinical records, underwriting files, RFPs and RFP responses.
What LLMs does it use?+
Multi-model by design. The four-tier engine routes across a semantic cache, a cheap open-weight model (Llama / Mistral) for standard queries, a mid-tier model (Claude Haiku / GPT-5 Mini) for complex extraction, and a premium model (Claude Sonnet 4.5 / GPT-5) only when the query justifies it.
How long does implementation take?+
Typical pilot: 2–4 weeks. Production rollout with SSO, audit trails, redaction and role-based access: 6–10 weeks. Reference clients are live in banking, insurance and legal.
Is Document Intelligence compliant with SOC 2, HIPAA, GDPR?+
Yes. Single-tenant deployment inside your compliance perimeter. SOC 2, HIPAA and GDPR-ready by architecture. On-prem and air-gapped deployments available for the most sensitive workloads.
Do I own my data?+
Yes. Full data ownership, zero third-party training, exportable at any time. Embeddings and vector indexes live in your infrastructure.
See it on your documents.
30 minutes with a senior Origami engineer. We'll show you the four-tier routing, page-level citations and sovereign deployment — and if you send us a sample document set under NDA, we'll run three of your real questions on it live.
- See your questions answered on your docs
- NDA signed before any document is shared
- Transparent cost model, per-answer
Prefer email? hello@origami.dev