Nutrient has made its Data Extraction API generally available, targeting one of the less glamorous but increasingly important problems in enterprise AI: turning messy documents into structured, traceable data that software agents can actually use. The API processes PDFs, scans, images and Office files, returning spatial JSON or Markdown and extracting schema-defined fields with page references, bounding boxes and source-grounding signals.
Large language models can summarize contracts, interpret invoices and answer questions about PDFs, but getting reliable information from those documents into an automated business process is a different challenge. A plausible answer is not necessarily an auditable one, particularly when an AI agent is making decisions that eventually affect payments, claims, approvals or compliance.
That is the gap Nutrient is targeting with the general availability of its Nutrient Data Extraction API, a document parsing and structured data extraction service designed for AI agents, retrieval-augmented generation (RAG) systems and enterprise automation.
The platform accepts PDFs, scanned documents, images and Office files. It can return Markdown for search, RAG and document question-answering workflows, or spatial JSON when downstream applications need information about layout, reading order, tables, forms and other page-level structures.
More importantly, Nutrient is positioning source grounding as part of the extraction layer rather than something developers have to bolt on afterward.
For extracted fields, the API can return page references, bounding boxes, source blocks, match labels and confidence signals. A field marked as a fuzzy match or not found can be routed to human review, while higher-confidence results can continue through an automated workflow.
That distinction matters as companies move from chatbot experiments toward AI agents that take actions.
McKinsey’s 2025 global AI survey found that 88% of respondents said their organizations regularly use AI in at least one business function, while 62% were at least experimenting with AI agents. Yet only about one-third said their organizations had begun scaling AI programs across the enterprise.
The implication is straightforward: enterprises are not short of AI models. They are still working out how to connect those models to dependable data, processes and controls.
Nutrient’s approach is to make document understanding configurable. Its API offers four processing modes—text, structure, understand and agentic—so developers can trade off processing depth, cost and speed depending on the document. The company says the system can identify tables, forms, formulas, charts, handwriting, checkboxes and headings, with OCR support for more than 100 languages.
That puts Nutrient into an increasingly crowded layer of the AI stack between raw documents and applications. Microsoft, Google and Amazon already provide cloud-based document intelligence capabilities, while specialist vendors and open-source projects such as Docling, LlamaParse and Unstructured compete for developers building RAG pipelines and AI applications.
Nutrient’s differentiation is less about claiming that an LLM can read a document and more about what happens after extraction.
The company publishes benchmarks based on the 200-document opendataloader-bench corpus. In its July 2026 results, Nutrient reports an overall accuracy score of 0.932 for its understand mode, with separate measurements for reading order, table structure and heading hierarchy. Those are Nutrient’s own benchmark results and should not be treated as an independent industry certification.
The company also reports that its understand and agentic modes scored 0.930 in its comparison against several open-source document processing systems. Its benchmark page says those tests were run on an Apple M3 Ultra using software versions current as of July 6, 2026.
Nutrient has also released an open grounding-en model through Hugging Face. The model is designed to assess whether a number, date or factual claim is actually supported by evidence in a source document. On Nutrient’s published evaluation, it achieved a 0.923 ROC-AUC score on number-grounding tasks, compared with lower scores for several general-purpose NLI models.
For enterprise AI teams, the broader trend is significant. AI infrastructure spending is moving rapidly from experimentation toward production: IDC estimates worldwide AI infrastructure spending reached $318 billion in 2025, more than double the previous year’s figure. Gartner, meanwhile, forecasts $2.59 trillion in worldwide AI spending for 2026 and expects AI infrastructure to account for more than 45% of that market.
That investment is feeding a much larger software ecosystem around AI development frameworks, cloud platforms, LLMs, agents and enterprise applications. But infrastructure alone does not solve the data-quality problem.
For document-heavy industries such as financial services, healthcare, insurance, legal services and government, the ability to show where an AI-generated value came from can be as important as the value itself.
Nutrient says its API is available through REST, with a free tier providing 5,000 monthly credits for new accounts. It also offers a visual Studio environment and browser-based extraction demos.
The bigger question is whether source-grounded extraction becomes a standard component of enterprise AI infrastructure. As organizations shift from asking AI to generate answers toward asking agents to execute workflows, traceability, exception handling and human review are becoming architectural requirements—not merely features.
Market Landscape
The enterprise document-AI market is shifting from OCR and PDF parsing toward AI-native document infrastructure. Traditional extraction tools focused on converting files into searchable text; newer systems increasingly preserve document structure, reason across layouts and feed structured outputs into RAG pipelines and autonomous workflows.
Nutrient competes in this layer with cloud services from Microsoft, Google and Amazon, alongside specialist platforms and open-source frameworks. The competitive battleground is increasingly moving beyond raw extraction accuracy toward grounding, latency, cost, observability, governance and integration with AI agents.
That evolution mirrors the wider enterprise AI market. Gartner forecasts AI models and platform spending of $64 billion in 2026, up 63.4% year over year, with spending increasingly focused on providers that can demonstrate reliability, cost efficiency and measurable outcomes.
Top Insights
- Nutrient’s API turns PDFs, scans and Office files into structured, source-grounded data designed for RAG pipelines and autonomous enterprise workflows.
- Source citations, page references and bounding boxes address a key weakness of probabilistic LLM extraction: limited traceability after an answer is generated.
- Four processing modes let developers balance extraction depth, speed and cost instead of applying the most expensive AI processing to every document.
- The platform enters a competitive market alongside Microsoft, Google, Amazon and specialist document-AI vendors targeting enterprise automation.
- The broader opportunity is production AI: moving document intelligence from prototypes into auditable workflows involving finance, healthcare, legal and insurance data.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
