Nutrient Data Extraction API launches for source-grounded, production-ready document AI

Nutrient Data Extraction transforms PDFs, scans, images, and Office files into structured document data — parsing content into spatial JSON or Markdown, and extracting schema-defined fields with per-field confidence signals and citations back to the source, helping enterprises operationalize agentic document workflows with reliable structured data.

RALEIGH, N.C., Sept. 9, 2026 /PRNewswire/ -- Nutrient Data Extraction API, a document parsing and structured data extraction service, is now generally available for enterprises building AI agents, RAG systems, large corpus document query, and enterprise automation. Nutrient Data Extraction helps organizations operationalize AI by transforming complex documents into reliable structured data for production-ready agentic workflows. 

Nutrient Data Extraction parses PDFs, scans, images, and Office files into spatial JSON or Markdown, and extracts schema-defined fields with per-field confidence signals and source citations, giving enterprises accurate structured data from documents enabling production-capable agentic workflows. 

Nutrient Data Extraction at a glance:

• Inputs: PDFs, scans, images, and Office files
  
  
• Parse outputs: spatial JSON or Markdown
  
  
• Modes: text, structure, understand, and agentic
  
  
• Structured extraction: customer-defined JSON Schema
  
  
• Extraction grounding: page references, bounding boxes, source blocks, match labels, and confidence signals
  
  
• Routing: fuzzy_match and not_found can go to human review
  
  
• Access: REST API, free tier, Studio, and getting-started guide
  
  
• Languages: 100+ for OCR
  
  
• Security: HTTPS/TLS encryption in transit, SOC 2 Type 2 audited infrastructure 

The reliability gap in agentic document workflows

AI has made document intelligence possible at enterprise scale. Documents can now be read, summarized, classified, extracted, and acted on in seconds. 

But generic LLM extraction is inherently probabilistic: the same document can produce different outputs under different conditions, lose structure, and return values with no traceable link back to the page they came from. Teams either accept the risk of unverified data flowing into ERPs, claims systems, and approval workflows, or they revert to manual review. 

Organizations heavily need data that is reliable, explainable, and traceable back to its source for agentic workflows to be trustworthy and ready for production. 

That's the reliability gap Nutrient Data Extraction was built to close. Most LLMs can extract data from documents. Far fewer systems can extract it accurately, consistently, and explain where it came from at a quality level that stands up in an audit.

That's the difference between model output alone and source-grounded document infrastructure built for production-ready agents. It's the difference between AI that's impressive in a demo and AI that can be relied upon by the human signing off on the business outcome.

And because Nutrient Data Extraction is a foundational part of Nutrient's broader document platform, teams can move from parsing and extraction to unlocking routing, governance, human review, and execution all without downstream workflow automation losing the auditable link back to the source document.

Built for multiple levels of understanding.

Teams, including Nutrient's implementation teams, are able to customize the output format per request based on the type and complexity of the document and the system requirements whether that's spatial JSON for layout-aware elements with reading order, bounding boxes, tables, and key-value regions, or Markdown for RAG, search indexing, and general document Q&A. Nutrient Data Extraction detects tables, forms, formulas, charts, handwriting, checkboxes, and headings, and supports more than 100 different OCR languages.

Not every document requires the same level of analysis and reasoning to get to accurate structured data. Nutrient Data Extraction lets teams optimize for speed, cost, or depth through four processing modes: text, structure, understand, and agentic, allowing processing to scale only when additional understanding is required.

Every extracted value traces back to the source

The extract endpoint maps a document to a customer-defined or intelligently generated JSON Schema and returns each requested field with a bounding box, page reference, the source blocks it was drawn from, a confidence signal, and a label describing how the value was grounded. Source grounding can drive workflow decisions automatically. Fields labeled "fuzzy_match" or "not_found" can be routed to human review, while high-confidence matches continue directly into downstream workflows.

Unlike tools that present confidence as a probability, Nutrient Data Extraction treats confidence as a relative signal. That distinction helps organizations structure and make better workflow decisions instead of over-interpreting a single score.

"Most agents can pull data out of a document. Almost none of them can prove where the data came from," said Jonathan Rhyne, co-founder and CEO of Nutrient. "Every field we return points back to the exact spot in the source, and when we can't find something, we say so instead of guessing. In production, 'Trust us' isn't an answer enterprises can take to their auditors or something accountable parties will rely on."

Published accuracy benchmarks

Nutrient publishes open accuracy benchmarks and updates the published results as new versions ship. In our latest test run, understand mode scored 0.932 overall accuracy on the publicly available 200-document opendataloader-bench corpus, across reading order, table structure, and heading hierarchy. Text and structure modes are scored on the public leaderboard, while understand and agentic modes were evaluated internally against the same corpus. The original open-source comparison was run on July 6, 2026 and the current results and methodology are maintained at nutrient.io/api/data-extraction-api/benchmarks. 

Separately, Nutrient has published two open grounding artifacts on Hugging Face: grounding-en, a model that scores whether an extracted value is supported by evidence in the source document, released under Apache-2.0; and the grounding benchmark dataset it is evaluated against, released under CC-BY-SA-4.0.

What it handles in practice

Nutrient publishes interactive extraction demos with no signup or API key required, where hovering an extracted field highlights the exact region of the source document it came from:

• Healthcare: Table-row extraction from a CMS-1500 claim form, including the dropout-red ink grid that defeats most scanners.
  
  
• Government and vital records: Signature block detection on a mixed handwritten and printed birth record, separating a printed name from the adjacent cursive signature.
  
  
• Mortgage and financial services: Value isolation in a dense three-column financial table where Monthly Income, Total Assets, and Total Expenses sit side by side under near-identical labels.
  
  
• Legal and contracts: Full multi-sentence narrative extraction from a free-text field in a small claims filing, plus cross-page extraction across three pages.
  
  
• AI, search, and RAG: An appraisal report decomposed into semantic blocks with spatial coordinates, so a retrieval pipeline can cite its exact source.

Availability

Nutrient Data Extraction is generally available via a REST API or can be built into custom solutions by our implementation teams. New accounts receive 5,000 Data Extraction API credits per month at no cost, with no credit card required. Signup is agent-compatible. 

Developers and agents can send a first request in minutes using the getting-started guide, or open the interactive demos in a browser with no account. Nutrient Data Extraction Studio provides a visual testing environment for evaluating documents before writing any code. 

All API communication uses HTTPS/TLS encryption, and the platform is SOC 2 Type 2 audited, with reports available under NDA. Processing-run retention varies by plan. Customers can delete runs and runs expire after the plan-specific retention period. On plans without internal data retention, documents are explicitly not retained for service improvement or model training. 

To view documentation, review the published benchmarks, or get a free API key, visit nutrient.io/api/data-extraction-api.

About Nutrient

Nutrient is the deterministic document platform organizations rely on to operationalize production capable agentic systems for document centric workflows. It combines reliable document processing infrastructure for agents, intelligent routing and governance, and interfaces for the human in the loop. Built on a decade of enterprise document expertise, Nutrient provides agentic document workflow solutions for thousands of organizations worldwide, including more than 15 percent of the Global 500, thousands of commercial businesses across 80 countries, and more than 130 public sector organizations in 24 countries. Backed by Insight Partners, Nutrient is headquartered in Raleigh, North Carolina, with offices in England, France, and Austria. Learn more at nutrient.io.

SOURCE Nutrient