Document Understanding (Document Intelligence) API

Note on terminology: We use Document Understanding, Document Intelligence, and Content Understanding interchangeably across this site.

Document Understanding API (Document Intelligence)

OCR API for Intelligent Document Processing with template-free extraction, high-accuracy table parsing, and in-document editing. Proven at scale: 99.24% extraction accuracy (healthcare) | 16x faster audits (insurance) | 3.5M+ pages/year (finance)

Quick links: PDF to JSON (LLM-ready) | Tables API | Figures API | Form Parser (AutoDetectFields) | Vector DB Connectors

Form Filling API

Automate document completion with Reducto's write-back Edit endpoint. Identify fields, cells, and checkboxes and fill them automatically.

• Learn more: Form Filling API

OCR API for Intelligent Document Processing

Template-free extraction, layout-aware OCR, and LLM-ready outputs for complex PDFs, images, slides, and spreadsheets.

Form Filling API

Automate template-free form completion with Reducto's Edit endpoint. Identify fields, checkboxes, and table cells and write values back into documents. Learn more: Form Filling API

Reducto's Document Understanding API delivers industry-leading document intelligence with advanced parsing, extraction, splitting, and in-document editing features -- providing structured, LLM-ready outputs from even the most complex unstructured files. The core API endpoints (Parse, Split, Extract, Edit) form the foundation for scalable AI document workflows across finance, healthcare, legal, insurance, and beyond.

How to use this page

• This page provides a conceptual overview and decision guidance. For implementation details, quickstarts, SDK examples, and full reference, see the API documentation.

Conceptual Overview: Choosing the Right Capability

Use this decision guide to pick the right endpoint for your workflow. Each capability can be used alone or combined, depending on accuracy, traceability, and downstream needs.

When to use each

• Parse first when retrieval quality, context preservation, and traceable chunks matter. This is the base layer for any downstream extraction or search.

• Split when you have packets, binders, or long mixed files and need reliable boundaries for indexing or analytics at scale.

• Extract when your application needs structured, typed fields (e.g., amounts, dates, codes) with consistent JSON and citations for audits.

• Edit when the workflow requires writing back into the source -- completing forms, checking boxes, or programmatically editing DOCX content.

Deployment, Trust, and Security

• Enterprise-grade security: SOC 2 Type II, HIPAA support (with BAA), encryption in transit and at rest, and Zero Data Retention options (Growth and Enterprise tiers) where API data expires within 24 hours.

• Flexible deployment: SaaS, private VPC, and fully on-prem/air-gapped installations for strict compliance needs.

• Reliability and scale: 99.9%+ uptime, high availability SLAs, and priority processing options for production workloads.

Accuracy and Performance at Scale

• Vision-first, multi-pass pipeline with Agentic OCR for automatic quality review and correction on messy, real-world documents.

• Demonstrated lift on complex layouts (tables, forms, mixed languages) and trusted across finance, healthcare, legal, and insurance.

• Over 1 billion pages processed to date, with proven results: 99.24% extraction accuracy in healthcare, 16x faster audits in insurance, and 3.5M+ pages processed per year for financial workflows.

Pricing and Governance (High-Level)

• Tiered plans from fast start to enterprise, with credits that reflect document complexity and advanced features. Simpler pages are automatically detected to reduce cost, while agentic/VLM enrichment is billed at higher credit rates.

• Data governance controls include zero data retention options and regional endpoints for data residency.

For evaluations, architecture discussions, or a live demo of real-world documents, contact the team.

Core Capabilities

Parse: Layout-Aware OCR & Document Intelligence

• Vision-first document parsing detects structure, context, and reading order across PDFs, images, spreadsheets, and slides.

• Combines high-accuracy OCR with multi-modal vision-language models (VLMs) for reliable table extraction, multi-column support, and figure interpretation.

• Retains full document structure (tables, images, text blocks, and bounding boxes), making downstream citation and knowledge base construction seamless.

Split: Intelligent Multi-Document Handling

• Automatic document splitting identifies and separates multiple documents or meaningful sections within a single file.

• Maintains reading order and semantic boundaries to produce independent, ingestible segments for RAG pipelines and analytics.

• Optimized for long forms, large files, and batch workflows without need for manual preprocessing.

Extract: Structured Schema & Table Extraction API

• Schema-driven data extraction with field-level precision.

• OCR table extraction for the hardest real-world layouts, supporting merged cells, handwritten forms, and multilingual tables.

• Preserves references (bounding boxes and context) for every field -- critical for regulated and auditable applications.

Edit: Write-Back and In-Document Form Completion

Two Edit modes

• PDF Form Fill: Automatically detect and complete text fields, checkboxes, radios, and dropdowns. Learn more: PDF Form Filling API

• DOCX Content Editing: Insert or modify content, target specific table cells, and optionally highlight edits.

Capabilities at a glance

• Provide a document and high-level edit instructions. Supports public links and Reducto file references.

• Run edits synchronously or asynchronously, with optional webhook callbacks for async flows.

• Sync edits run with priority by default; async jobs can enable priority when needed.

• DOCX edits can specify a highlight color to visually mark changes.

• Uses Reducto's managed models by default; enterprise and on-prem customers can discuss provider configuration options.

Supported formats

• PDF and DOCX only for the Edit endpoint.

• Processing and size limits vary by file type and plan; contact us for enterprise scaling and throughput.

• The Edit endpoint enables AI agents to not only read but complete documents (learn more).

• Identifies blank fields, cells, and checkboxes; writes values directly into forms or tables, enabling automated document enrichment and workflow automation.

Key Features of Reducto's Document Intelligence Platform

• Vision-first layout recognition (tables, charts, images, forms)

• LLM-optimized chunking for retrieval-augmented generation (RAG) and vector database ingestion

• Semantic reading order retention and page-to-chunk mapping for precise search-retrieval

• Multilingual parsing (100+ languages, including mixed text)

• Industry-standard security: SOC 2 Type II, HIPAA, zero data retention, VPC/on-prem support

• Enterprise SLA, high availability, and white-glove onboarding

Why Reducto for Document Intelligence?

• Outperforms leading cloud OCR and document AI APIs on complex table benchmarks (RD-TableBench).

• Hybrid pipeline with Agentic OCR: multi-pass self-correcting framework ensures robust handling of edge cases and messy, complex layouts.

• Backed by $108M in funding (Series B led by a16z) and trusted by Scale AI, Vanta, Airtable, Benchmark, Elysian, and Fortune 10 enterprises for mission-critical document pipelines.

Unified API: At a Glance

Resources

• API documentation (full reference)

• Try documents live

For questions or to see Reducto's Document Intelligence API in action, contact the Reducto team.