Cinnamon AI

Flax Scanner reads a form without coordinate definitions, pre-training or a template build. That changes the economics of small document sets: a batch of 40 supplier invoices no longer needs a two-week setup before the first page is read. The same engine covers fixed layouts and irregular ones.

The scanner targets the pages that break older AI-OCR tools, including handwritten entries, merged cells, empty cells and star-mark grids. For finance teams, that's the difference between scanning half of a monthly batch and scanning all of it. Accuracy still varies by document quality, so plan a review step.

Cinnamon AI pairs its own IDP engine with large language models, so extraction can follow an instruction rather than a rigid field map. Users describe what they need from a document and the system returns the values. The practical gain shows up on mixed packets, where an invoice, a delivery note and a purchase order arrive in one PDF.

Super RAG keeps the layout and structure of a document when it indexes it, which helps with charts, figures and tables that plain text chunking destroys. Answers draw on those structures instead of isolated sentences. The vendor's NAACL 2025 industry paper describes the layout-aware graph approach behind it.

A single question often needs facts from several files. Super RAG runs multi-hop queries that search and reason across documents before answering. In a contracts team, that means one question can span the master agreement, an amendment and a price schedule without the user opening three folders.

Every answer can display the document and passage it came from, which makes verification fast. Deployments are single-tenant, either in Cinnamon AI's cloud or the customer's Azure environment, and admins can split user groups so each group only reaches its own files. The company holds Privacy Mark and ISMS certification.