Cinnamon AI — IDP Software Vendor

6 min read

On This Page

Tokyo-based AI company offering the Flax Scanner AI-OCR platform and Super RAG document LLM for financial appraisal, logistics, and enterprise knowledge management.

Cinnamon AI

$41.34MTotal funding raised
92.99%Character accuracy on invoices
~3 secPer-document processing speed
119Employees across 4 continents

Overview

Founded in 2012 in Tokyo, Cinnamon AI has evolved from document processing into a two-product intelligent document processing (IDP) stack. Flax Scanner HUB handles ingestion and extraction across unstructured forms. Super RAG, launched in March 2024, handles downstream retrieval and analysis of extracted content using a document-specialized large language model (LLM). Together they position the company as a full-stack IDP vendor rather than a point-solution OCR or retrieval-augmented generation (RAG) provider.

The company serves over 50 enterprise customers, including Toyota, Sumitomo Pharma, Fujitsu, and Daikin, across four continents. Total funding stands at $41.34M, according to CB Insights, with investors spanning financial services (Dai-ichi Life, Sumitomo Mitsui Banking), office equipment (OKAMURA, TRUSCO), and venture capital (Plug and Play Japan). CB Insights has included Cinnamon AI in its AI 100 list from 2018 through 2025 and in its Robotic Process Automation Expert Collections.

In early 2026, CEO Miki Hirano articulated a strategic shift toward "process redesign": prioritizing which business processes need AI over what the technology can do. The shift emphasizes specialized business agents with limited, well-defined objectives. Hirano's appointments to Japan's Growth Strategy Council (November 2025), AI and Semiconductor Working Group (December 2025), and Drug Discovery Working Group (December 2025) embed the company within national AI policy development.

By February 2026, that strategy was visible in go-to-market moves. Super RAG expanded to three pricing tiers, adding a starter plan that opens the product to smaller buyers beyond the enterprise accounts it originally targeted. A new version of Flax Scanner HUB was also released, though no version string or benchmarks are publicly available. The company exhibited at AI Expo Osaka in January, Life-Work Balance EXPO Tokyo in February, and Japan DX Week Nagoya in late February, signaling a deliberate regional coverage push within Japan. At Osaka, the company actively recruited system integrators and BPO companies as reseller partners, suggesting an indirect channel is being built alongside direct enterprise sales.

All product performance claims in this profile are vendor-stated unless otherwise noted. No third-party analyst coverage or independent benchmarks are publicly available as of April 2026.

How Cinnamon AI processes documents

Cinnamon AI's document processing architecture separates ingestion from retrieval across two products, with vertical-specific agents sitting above both.

Flax Scanner HUB handles the extraction layer using a three-engine architecture: coordinate-defined, feature-learning, and generative AI extraction. The system selects the optimal method per document type automatically. This distinguishes it from conventional coordinate-only OCR systems. The feature-learning engine requires no template creation, operating coordinate-free across industry documents, technical drawings, and photographic images. Processing speed is approximately 3 seconds per document. A no-code semi-automatic learning function allows teams to improve extraction accuracy without developer involvement.

Vendor-reported accuracy on financial documents: 92.99% character accuracy and 86.17% item-specific accuracy across 27 invoice items. On trade documents: 91% accuracy across 54 Commercial Invoice items, with automatic extraction of 50 Bill of Lading items.

Super RAG handles the downstream layer. Rather than splitting documents by character count or fixed token windows, Super RAG uses layout analysis to auto-detect document structure and chunk content by semantic boundaries. According to CB Insights, this approach addresses a known limitation of generic RAG implementations applied to structured documents. The system handles complex Japanese document types including dense tables, bar graphs, diagrams, and handwritten content while suppressing hallucinations. Three proprietary technologies underpin it: document analysis, knowledge injection, and automatic prompt generation. Cinnamon AI states: "By utilizing Super RAG, it is possible to extract information from complex forms including complex tables, charts, bar graphs, diagrams, and handwritten documents while suppressing hallucinations and supporting the use of all kinds of documents when using LLM."

Microsoft SharePoint integration is formally released. Box integration is planned for a future release. Both signal a push into Western enterprise workflows beyond the domestic Japanese market where the company has its strongest footprint.

Financial Appraisal AI Agents sit above both layers, automating first and second appraisal processes in financial services with human oversight reserved for complex later-stage evaluations. These agents represent the company's clearest move into vertical-specific automation beyond document extraction.

Use cases

Financial services

Financial institutions deploy Cinnamon AI's Financial Appraisal AI Agents for automated first and second appraisal processes, with human oversight reserved for complex evaluations. Finance teams also use Flax Scanner HUB for invoice processing, achieving 92.99% character accuracy and 86.17% item-specific accuracy across 27 invoice items without template configuration.

Logistics and trade

Logistics departments process shipping documents with 91% accuracy across 54 Commercial Invoice items and extract 50 Bill of Lading items automatically, without requiring coordinate-based templates.

Engineering and knowledge management

Engineering teams deploy Super RAG against technical documentation repositories. The system learns from query patterns over time, improving retrieval accuracy on incident prediction workflows, equipment manuals, and tacit-knowledge capture. This use case extends Cinnamon AI's footprint from document extraction into ongoing knowledge management, addressing the roughly 80% of enterprise data held in unstructured formats.

Pharmaceutical and life sciences

CEO Hirano's appointment to Japan's Drug Discovery Working Group (December 2025) signals emerging engagement with pharmaceutical document workflows, though no product deployments in this vertical are publicly documented as of April 2026.

Technical specifications

Feature Specification
Core products Flax Scanner HUB, Super RAG, Financial Appraisal AI Agents
Extraction engines Coordinate-defined, feature-learning, generative AI (auto-selected per document type)
Template requirements None (coordinate-free extraction)
Processing speed ~3 seconds per document
Invoice accuracy 92.99% character, 86.17% item-specific (27 items)
Trade document accuracy 91% (Commercial Invoice, 54 items); 50 Bill of Lading items auto-extracted
Super RAG chunking Semantic boundary detection via layout analysis (not character-count-based)
Super RAG pricing tiers Three tiers including starter plan (figures not disclosed)
Deployment options Multi-tenant SaaS, single tenant, on-premise, private cloud
Enterprise integrations Microsoft SharePoint (released); Box (planned)
Learning function No-code semi-automatic learning (Flax Scanner HUB); self-improving retrieval (Super RAG)
Third-party benchmarks None publicly available as of April 2026

Independent verification gap: All accuracy figures on this page are vendor-reported. No third-party benchmark or analyst assessment of Cinnamon AI's extraction accuracy has been published as of April 2026. Evaluators should request independent test results during procurement.

Company information

Headquarters: Tokyo, Japan (offices in US, Vietnam, Taiwan)

Founded: 2012

Founders: Miku Hirano (CEO), Hajime Hotta, Hiroaki Kitano, Mori Aki, Yoshiaki Ieda

Employees: ~119 across 4 continents

Funding: $41.34M total raised (per CB Insights), including Corporate Minority - III round; investors include Dai-ichi Life, Sumitomo Mitsui Banking, OKAMURA, TRUSCO, and Plug and Play Japan

CEO Hirano holds advisory positions on Japan's Growth Strategy Council and two working groups (AI and Semiconductors; Drug Discovery) as of late 2025. These appointments give the company visibility into national AI procurement and regulatory direction, which may benefit enterprise sales cycles with Japanese government-adjacent buyers.

Resources

For context on the OCR and RAG capabilities underlying these products, see OCR technology and agentic document processing. For Japanese-market alternatives with comparable OCR depth, see Cogent Labs.