RAG Frameworks · NanoNets

docext

docext is an on-premises document processing toolkit that extracts structured data from PDFs and images using vision-language models, without requiring OCR. It converts documents to markdown, extracts key information from invoices and passports, and includes a benchmarking platform to compare model performance on document intelligence tasks.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/NanoNets/docext

GitHub stars

148

Forks

Python

Primary language

Apache-2.0

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	NanoNets/docext
Owner	NanoNets
Primary language	Python
License	Apache-2.0 — OSI-approved
Stars	2k
Forks	148
Open issues	21
Latest release	v0.1.14 (2025-06-30)
Last updated	2026-03-17
Source	https://github.com/NanoNets/docext

What docext is

Python-based toolkit leveraging VLMs for OCR-free document understanding, featuring PDF/image-to-markdown conversion with LaTeX/signature/watermark detection, structured field extraction with confidence scoring, REST API, and a public leaderboard benchmarking suite covering KIE, VQA, table extraction, and document classification across multiple models (Claude, Gemini, Qwen, etc.).

Quickstart

Get the docext source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/NanoNets/docext.git2cd docext3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Invoice and Receipt Processing

Extract line items, amounts, dates, and vendor information from invoices and receipts with confidence scores. Pre-built templates and multi-page support enable batch processing on-premises without cloud dependencies.

Identity Document Verification

Structured extraction from passports, driver's licenses, and visas with field confidence scoring. On-premises deployment maintains data sovereignty for sensitive personal documents.

Compliance and Document Classification

Batch categorize incoming documents (contracts, permits, forms) and extract compliance-relevant fields. Benchmark different VLMs before deployment to optimize cost and accuracy trade-offs for your document mix.

Implementation considerations

VLM selection impacts cost, latency, and accuracy—use the public leaderboard (idp-leaderboard.org) to benchmark against your document types before committing to a model backend.
Infrastructure must support VLM inference; pulling models like Nanonets-OCR-s (3B parameters) or larger requires adequate GPU/CPU and memory.
Custom field definitions and template creation require domain expertise; pre-built templates (invoice, passport) provide a starting point but custom schemas need careful design for confidence scoring reliability.
Confidence scores are model-dependent; validate calibration on your document distribution to avoid false confidence in edge cases.
REST API deployment requires standard DevOps practices (containerization, scaling, monitoring) not documented in the excerpt.

When to avoid it — and what to weigh

Heavy real-time, low-latency requirements — VLM inference latency is typically seconds per document. Not suitable for millisecond SLAs in synchronous workflows.
Handwriting-heavy or heavily degraded documents — While the toolkit claims signature and watermark detection, performance on poor-quality or entirely handwritten documents is not benchmarked in the provided data.
Windows-only or restricted Linux environments — README states Linux and macOS support only. Deployment in Windows-locked enterprises requires validation.
Need for proprietary, fine-tuned model weights — Toolkit uses public VLMs (Claude, Gemini, Qwen). If you require proprietary model IP, this open-source approach may not fit.

License & commercial use

Apache License 2.0 (OSI-approved permissive license). Permits commercial use, modification, and distribution under the same license terms with liability and warranty disclaimers.

Apache 2.0 permits commercial use without explicit vendor permission. However, commercial deployment of VLM inference (Claude, Gemini) requires separate licensing and API agreements with those providers. Nanonets-OCR-s model availability and commercial terms should be confirmed independently. Verify your use case with legal counsel.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	High
DEV.co fit	Good
Assessment confidence	Medium

Security considerations

On-premises deployment avoids cloud-side data exposure if infrastructure is properly secured. No explicit security audit, vulnerability disclosure policy, or supply-chain provenance data provided. VLM API keys (for Claude/Gemini fallback) require secure secret management. Input validation and output sanitization practices not documented.

Alternatives to consider

Unstructured.io

Modular document parsing library with on-premises LLM support; however, less integrated benchmarking and fewer pre-built templates than docext.

Hugging Face Document AI ecosystem (LayoutLM, LayoutXLM)

Fine-tunable transformer models for document understanding; requires more ML ops expertise and custom pipelines, but no vendor lock-in to specific VLM providers.

Amazon Textract or Azure Document Intelligence

Managed cloud services with built-in OCR, KIE, and table extraction; simpler API but cloud-dependent, higher per-page cost, and no on-premises option.

Software development agency

Build on docext with DEV.co software developers

Evaluate docext against your document types using the public leaderboard, then engage our team to design a scalable, secure on-premises deployment with custom templates and confidence calibration.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

opendataloader-pdf PageIndex knowhere

Related on DEV.co

Explore the category and the services that help you build with it.

RAG Frameworks AI Development API Development DevOps & Deployment Talk to DEV.co

docext FAQ

Can I run docext entirely offline without calling external VLM APIs?

Partially. Nanonets-OCR-s (3B model) and some open VLMs can be self-hosted, but the leaderboard and pre-built templates reference Claude/Gemini. Confirm which models support offline inference for your use case.

What is the performance/cost difference between local VLMs and cloud APIs?

Not provided in documentation. Use the public leaderboard to compare accuracy metrics across models, then benchmark latency and cost in your environment.

Is there a managed SaaS version of docext, or is it self-hosted only?

Documentation indicates on-premises deployment. Nanonets (the company) offers commercial services; check their website for managed offerings.

How do I handle multi-language documents?

Not explicitly documented. Depends on the underlying VLM's language support. Benchmark via the leaderboard or test with your documents.

Software developers & web developers for hire

DEV.co helps companies turn open-source tools like docext into production software. Our software development services cover the full lifecycle — architecture, web development, integration, and maintenance — delivered by software developers and web developers who ship. Engage our software development agency to implement or customize it for your rag frameworks stack.

Ready to Deploy docext?

Evaluate docext against your document types using the public leaderboard, then engage our team to design a scalable, secure on-premises deployment with custom templates and confidence calibration.

Talk to DEV.co Schedule a Consultation