DOCUMENT AI

Document infrastructure for agents

Structure documents for coding agents and AI harnesses like Codex, Claude Code, Claude Cowork, OpenClaw, and custom MCP clients. Give them reliable document tools they can call with 100 percent reproducible, auditable results — in your application, under your control.

Document AI

Why Nutrient document AI

Other vendors either prepare documents for AI or layer chatbots on top. Nutrient spans the full pipeline — infrastructure, intelligence, and governed agents in one integration.

Full pipeline

Improve input quality, extract structure and content, reason through tasks with AI, edit and transform documents, and then deliver accessible, compliant output with audit-ready data. One platform covers the entire journey.

AI-grade document engine

Generic AI can interact with documents — but the quality of its output depends on the document engine underneath. Twenty years of document technology means AI gets the best tools, the highest-quality data, and the most reliable structure to act on.

Governed document agents

Your AI agent can think about documents. Nutrient gives it governed tools to extract, fill, annotate, redact, compare, sign, and produce outputs with the approval flow you define.

Your model, your infrastructure

Bring any LLM — OpenAI, Anthropic, AWS Bedrock — or run open source locally. Deploy the platform on-premises, in managed cloud, or in-browser. You choose the model, and you choose where it runs.

The Nutrient document AI pipeline

Three phases. One platform. Every step uses purpose-built technology — AI where it adds value, and deterministic operations where precision is required.


Structure: AI-ready documents

Before the model: Documents need to be clean, structured, and machine-readable. Nutrient converts, extracts, and structures documents so AI can consume them reliably.


  • Clean, structured inputs for any LLM
  • Accurate extraction with confidence scores
  • 100+ formats converted and normalized
Convert and normalize

Turn 100+ file formats — PDFs, Office files, scans, images — into clean Markdown, HTML, or structured JSON that any model can consume.

Extract structure and content

Pull tables, key-value pairs, form fields, and semantic layout from unstructured inputs — using hybrid AI that combines local models with vision language models.

Classify and route

Automatically categorize documents by type — invoices, contracts, IDs, receipts — and extract the data you need using natural language descriptions, not templates or regex.


Intelligence: Document-ready AI

AI agents that understand documents and take action on them — governed by your rules, embedded in your application.


  • Agents that modify documents, not just read them
  • Configurable autonomy with human approval where needed
  • Reliable document tools your agent can call to get reproducible, auditable results every time
AI Assistant editing agent

Autonomous multistep workflows — extraction, form filling, annotation, redaction — through natural language.

Permanent AI redaction

Irrecoverable PII removal — staged by AI, approved by humans, and executed by a battle-tested redaction engine.

Semantic document comparison

AI-enhanced diffs that surface meaning-level changes, not just text differences.


Output: Compliant results

Every AI action surfaces in the same document viewer users already work in — with custom approval workflows before anything is finalized. Then, documents come out digitally certified, accessible, and ready for long-term preservation.


  • Human review and approval in the viewer
  • Certified, accessible, archival-ready output
  • Deterministic artifacts with clear provenance
Certify and make accessible

PAdES-compliant digital signatures make documents legally binding. PDF/UA auto-tagging makes them accessible. These are the finalization steps that generic AI platforms skip entirely.

Archive for compliance

Convert to PDF/A for long-term archiving that ensures documents remain readable, verifiable, and regulation-ready for decades. Signed, sealed, preserved.

Deterministic, audit-ready output

Turn agent activity into final outputs that are reproducible every time — with human verification, clear provenance, and document artifacts that can be archived, audited, and relied on downstream.


COMPARE

One pipeline, multiple entry points

Nutrient document AI adapts to how your team works — whether you’re embedding into an app, connecting AI agents, or automating across business systems.

SDKs
DWS MCP Server
Workflow Automation
Ideal for
Product and engineering teams
Developers building AI agents
Operations and process owners
What you get
Embed the full pipeline — AI-powered extraction, governed agents, redaction, and compliant output — in any web, mobile, or desktop app.
Give AI agents reliable document operations — convert, sign, redact, extract — via natural language through MCP-compatible clients.
AI-powered data extraction and document classification — drag and drop, no code required.
Pipeline coverage
Structure + intelligence + output — full pipeline with AI Assistant, document generation, and compliance output.
Agent-ready operations — deterministic document tools any AI framework can call reliably.
Automated extraction and classification at scale — reduce errors, rework, and manual effort.

Works with the agent tools your team already uses

Start with the integration layer that matches your stack — MCP for agent frameworks; OpenClaw for native tool-based workflows; or the CLI for scripts, jobs, and CI.

MCP Server

Connect document operations to Claude, GPT, and other agent frameworks through a standard MCP interface. Let agents convert, extract, sign, redact, and generate documents through natural language without building custom glue code first.

OpenClaw plugin

Run Nutrient document operations as native tools inside OpenClaw-based agent workflows. Give agents direct access to extraction, OCR, conversion, redaction, signing, and other document actions inside a governed tool runtime.

DWS CLI

Automate conversions, extraction, OCR, signing, and output generation from scripts, batch jobs, and CI pipelines — ideal when you want agent-compatible document workflows without embedding a full SDK or standing up custom orchestration first.

DEPLOYMENT OPTIONS

Deploy anywhere, keep your model options open

The pipeline adapts to your infrastructure — not the other way around. Flexible deployment to fit your governance, latency, and scale requirements.

In-browser (WebAssembly)

Embed document viewing, annotation, and editing directly in the frontend with no backend. Add AI capabilities by connecting to the AI Assistant backend service.


Self-hosted

Run the full pipeline on your infrastructure. Keep sensitive documents and AI processing entirely within your perimeter.


Managed cloud

The fastest path to production, with zero DevOps. Nutrient manages your environment with dedicated infrastructure and SLAs.


Shared cloud

Quick scale without complexity. Simple setup, no infrastructure required.


Use the model that matches your governance, cost, latency, and reasoning needs — without changing the document infrastructure underneath.

Built for security, designed for ownership

Your documents, your models, your infrastructure. Nutrient keeps you in control.

Deploy on your terms

Self-hosted LLMs keep sensitive documents entirely within your infrastructure. No data leaves your perimeter.

Cloud LLM safety

When using third-party models, Nutrient ensures your data is never used for training.

AI usage transparency

Nutrient never trains its core models on your documents. Your data stays yours.

Verified security

SOC 2 Type 2 audited, and GDPR- and CSA-compliant — trusted by startups and Global 500 enterprises.

Free 30-day trial

See Nutrient document AI in your application today.

Frequently asked questions

What is Nutrient document AI?

Nutrient document AI is the complete document intelligence pipeline — from structured extraction, through governed AI agents to compliant output. It structures documents for AI (conversion, OCR, Vision API, AI Document Processing), embeds AI that acts on documents (AI Assistant with chat and editing agents, MCP Server), and produces compliant output (PDF/A archiving, PDF/UA accessibility, digital signatures, document generation). All of this is embedded in your application and under your control.

What makes Nutrient different from other document AI solutions?

Most vendors do one of two things: prepare documents for AI (infrastructure only) or bolt chatbots onto documents (AI only). Nutrient spans the full pipeline:

  • Structure — Convert, extract, and normalize documents so AI can consume them reliably. Vision API, OCR/ICR, and AI Document Processing handle everything from scanned invoices to complex blueprints.
  • Intelligence — Embedded AI agents that don’t just read documents but modify them — extraction, form filling, annotation, redaction — through natural language, governed by your rules.
  • Output — Turn AI outputs into production-ready documents with PDF/A archiving, PDF/UA accessibility auto-tagging, digital signatures, and generation from templates.

Twenty years of document expertise makes the AI layer trustworthy. Any company can plug an LLM into a viewer. Making it reliable at enterprise scale requires deep domain knowledge.

What is AI Assistant and what can the editing agent do?

AI Assistant includes two purpose-built agents embedded in our Web, iOS, Android, and React Native SDKs:

  • Document editing agent — The differentiator. It autonomously plans and executes multistep document workflows: extracting structured data, filling forms, adding annotations, and redacting sensitive content — all through natural language. It selects tools, verifies outcomes, and adapts. Organizations control what runs automatically, what requires approval, and what is blocked through three-tier governance.
  • Chat agent — Fast Q&A, summarization, and translation. Read-only, cost-efficient, low-latency. Baseline capability across all platforms.

Both agents connect to the LLM provider of your choice — OpenAI, Anthropic, Azure, AWS Bedrock, or self-hosted models.

What is the MCP Server and how does it work with AI agents?

The MCP Server gives any AI agent framework — Claude, GPT, LangGraph, or custom — access to deterministic document operations through natural language. Hand the agent a prompt, get a compliant PDF back.

  • Operations available — Convert, sign, redact, merge, OCR, archive (PDF/A), make accessible (PDF/UA), extract structured data.
  • Sandboxed execution — Every operation runs in an isolated workspace. No access to file systems, email, or anything beyond the document.
  • Designed for AI discovery — Deterministic JSON schemas, llms.txt metadata, and MCP-native interfaces so agents can find and use operations without human intervention.

Open source, one-line install, Docker-friendly.

Which LLMs are supported, and can we use self-hosted models?

Nutrient document AI supports any LLM provider:

  • Cloud providers — OpenAI, Azure OpenAI, AWS Bedrock, Anthropic Claude.
  • Self-hosted — Any model compatible with the OpenAI API specification, including open source models like Llama via vLLM or Ollama.

You choose based on data residency, cost, latency, or existing cloud relationships. Swap providers with one line of configuration.

How does Nutrient handle data privacy, security, and compliance?
  • Self-hosted deployment — Run the full pipeline on your infrastructure. Documents and AI processing stay entirely within your perimeter.
  • Cloud LLM policies — When using third-party models, Nutrient configures connections to prevent your data from being used for training.
  • No training on your data — Nutrient never trains its core models on customer documents or AI-generated content.
  • Verified security — SOC 2 Type 2 audited, and GDPR- and CSA-compliant.
  • Governed AI agents — Three-tier autonomy controls (autonomous/confirmation-required/prohibited) ensure AI actions match your compliance requirements.
What document formats and types does the pipeline support?

The pipeline handles 100+ file formats:

  • Input — PDF, DOCX, XLSX, PPTX, images (JPEG, PNG, TIFF), scanned documents, HTML, and more.
  • Extraction — Tables, key-value pairs, form fields, and semantic structure from both digital and scanned documents via OCR, ICR, and Vision API.
  • Output — PDF, PDF/A (archival), PDF/UA (accessible), HTML, Markdown, JSON, and generated documents from templates.
Can Nutrient handle scanned documents and complex layouts?

Yes. Vision API combines local AI models with vision language models to extract structured data from scanned documents, handwritten notes, and complex layouts — including tables with merged cells, multicolumn text, blueprints, and forms. OCR handles character recognition, while ICR adds layout understanding and semantic classification. Every extraction returns JSON with bounding boxes and confidence scores.

Who uses Nutrient document AI?

Nutrient is trusted by organizations across regulated, document-heavy industries:

  • Legal — Harvey, the $8B+ legal AI platform, chose Nutrient as its document infrastructure, scaling more than 50 percent month over month.
  • Financial services — Invoice processing, KYC verification, expense automation, and PCI-compliant document handling.
  • Government — 130+ public sector organizations across 24 countries rely on Nutrient for document processing.
  • Healthcare — Patient intake, claims processing, PHI redaction, and lab report extraction.
  • Aviation — IBM’s Pilotbrief serves 34,000+ commercial pilots on Nutrient-powered document infrastructure.
How is Nutrient document AI priced?

Pricing depends on deployment model (cloud vs. self-hosted), scale (users, document volume), and which pipeline capabilities you need. AI Assistant connects to your own LLM provider — you pay Nutrient for the platform and your LLM provider for inference. A built-in cost calculator estimates per-document AI costs. For a personalized quote, contact our Sales team.