Solutions · Document Intelligence · Chat with PDF

Ask your documents. Get grounded answers.

The most advanced local document Q&A technology available. Query PDFs, Office documents, HTML, Markdown, and more. Powered by semantic RAG, adaptive layout analysis, and a fully customizable agent with tool calling and memory. 100% on-device.

Start building free View complete demo

Semantic RAG Adaptive processing Layout analysis

Adaptive engine

Auto-selects OCR, VLM, or text extraction based on content.

Layout analysis

Full document structure understanding with page elements.

Semantic RAG

Embedding-based retrieval with intelligent chunking.

Source attribution

Every answer traced to document, page, and passage.

0
Cloud calls

100%
Traceable

∞
Documents

How it works

Semantic RAG meets document layout analysis.

LM-Kit.NET delivers the most advanced local document Q&A technology available. The underlying system combines semantic retrieval-augmented generation with a complete document layout analysis stack. Supports PDF, Office documents (Word, Excel, PowerPoint), HTML, Markdown, images, and more.

The engine is extremely fast, using an adaptive approach that intelligently engages OCR, Vision Language Models, layout processing, or direct text extraction based on content discovery. Each page is analyzed and processed using the optimal strategy automatically.

Built by IDP pioneers: This isn't a wrapper around generic RAG. It's purpose-built document intelligence from a team with 20+ years of experience processing billions of documents in production worldwide.

Program.cs

using LMKit.Retrieval;
using LMKit.Model;

// Load your preferred models
var chatModel  = LM.LoadFromModelID("your-chat-model");
var embedModel = LM.LoadFromModelID("your-embedding-model");

// Create document chat instance
using var chat = new PdfChat(chatModel, embedModel);

// Load documents - adaptive processing kicks in
await chat.LoadDocumentAsync("contract.pdf");
await chat.LoadDocumentAsync("scanned-report.pdf");

// Ask questions in natural language
var response = await chat.SubmitAsync(
    "What are the payment terms?");

// Get answer with source references
Console.WriteLine(response.Response.Completion);
foreach (var src in response.SourceReferences)
    Console.WriteLine($"  → {src.Name}, p.{src.PageNumber}");

Under the hood

Adaptive document processing pipeline.

Content-aware processing that automatically selects the optimal extraction strategy for each page based on content discovery.

Step 01

Document import

PDF, Office, HTML, Markdown, images analyzed page-by-page.

Step 02

Auto-detection

Content type determines processing mode.

Step 03

Extraction

Text, OCR, VLM, or layout analysis.

Step 04

Semantic RAG

Embeddings, retrieval, grounded answers.

Processing modes

Intelligent content-aware processing.

The engine automatically selects the optimal strategy for each page based on content analysis, or you can specify your preference.

Default · Auto

Content-driven selection

Analyzes each page and selects the best processing strategy automatically. Uses VLM for image-heavy pages, text extraction for digital content.

Zero configuration required
Optimal quality/speed balance
Handles mixed document types

Mode · TextExtraction

Fast, direct processing

Extracts text directly from PDF structure with OCR fallback for scanned or image-based pages. Maximum speed.

Fastest processing
Lower resource usage
Best for clean digital PDFs

Mode · DocumentUnderstanding

VLM-powered analysis

Vision Language Models analyze pages visually to understand layout, structure, tables, and relationships. Markdown output.

Best for complex layouts
Preserves document structure
Tables, forms, mixed content

Core capabilities

Production-ready document intelligence.

Everything you need to build document Q&A applications that actually work.

Capability

Multi-document queries

Load multiple documents and ask questions that span all of them. Compare contracts, cross-reference reports, search collections.

Capability

Source attribution

Every answer includes document names, page numbers, and custom metadata. Full traceability for compliance and audit.

Capability

Intelligent caching

Processed documents cached via IVectorStore. Subsequent loads are instant. Filesystem or custom backends (Qdrant, PostgreSQL).

Capability

Smart context

Small documents included in full. Large documents use semantic passage retrieval. Automatic optimization per document.

Capability

Multi-turn dialogue

Follow-up questions maintain context. Natural conversation flow. Ask clarifying questions without re-explaining.

Capability

100% local

All processing on your infrastructure. Documents never leave. Air-gapped deployments. HIPAA, GDPR, compliance-ready.

Capability

Agentic capabilities

PdfChat is a fully customizable document agent. Connect external tools, maintain conversation memory, and integrate with MCP servers for extended functionality.

Try it now

Complete demo application.

A fully-featured console application demonstrating all capabilities, ready to run.

Featured demo

Chat with PDF demo

Interactive console app that lets you load PDFs, ask questions, and see the full document Q&A pipeline in action with source references, generation stats, and real-time streaming.

Model selection with download progress
Standard or vision processing modes
Multi-document loading with caching
Interactive commands (/help, /status, /add)
Token counts and generation speed metrics

View sample guide GitHub source code

terminal

$ dotnet run
# Select model (0-8 or custom URI): 2
# Loading model... ████████ 100%
# Processing mode: 1 (Vision)
# Enter PDF path: report.pdf
# ✓ Loaded: 45 pages, passage retrieval

You: What was the Q4 revenue?
# Retrieved 5 passages in 23ms

Assistant: Based on the Q4 financial report,
revenue was $4.2M, representing 15% YoY growth...

# → Source: report.pdf, Page 12
# Tokens: 156 | Speed: 42.3 tok/s

Use cases

Built for real-world applications.

Document intelligence that solves actual business problems.

Use case

Contract analysis

Query legal agreements for specific clauses, obligations, termination conditions, and payment terms with full source attribution.

Use case

Financial review

Ask questions about revenue, expenses, projections, and risk factors across multiple financial reports and statements.

Use case

Technical documentation

Search manuals and specifications for configuration details, procedures, system requirements, and troubleshooting steps.

Use case

Research & academia

Query research papers for methodology, findings, citations. Cross-reference multiple sources for literature reviews.

Use case

Compliance & audit

Verify policy adherence with traceable source references. Generate audit trails with document and page attribution.

Use case

Customer support

Build knowledge bases from product documentation. Answer customer questions automatically with grounded responses.

Choose your models

LM-Kit.NET supports a wide range of vision-capable chat models, embedding models, and specialized OCR models. Browse our model catalog to find the right combination for your use case, hardware, and accuracy requirements.

Browse model catalog

API reference

Key classes.

The building blocks for document Q&A applications.

Class

`PdfChat`

High-level document agent for question-answering. Supports tool calling, conversation memory, MCP integration, and full customization.

View documentation

Class

`DocumentRag`

Lower-level document RAG engine with full control over processing modes, chunking, and retrieval parameters.

View documentation

Interface

`IVectorStore`

Interface for embedding storage and caching. Use FileSystemVectorStore or implement custom backends.

View documentation

Class

`VlmOcr`

Vision-based document parser using VLMs. Preserves layout and structure as markdown output.

View documentation

Related capabilities

Chat plus the rest of Document Intelligence.

OCR

When the PDF is a scan, OCR runs transparently. Native engine plus VLM OCR with PaddleOCR-VL, GLM-OCR, LightOnOCR.

OCR page

Document to Markdown

Universal converter that picks the right strategy per page. The same primitive feeds the chat pipeline.

Markdown converter

Document RAG engine

Lower-level control: explicit lifecycle, chunking strategies, custom vector stores, source attribution.

Document RAG

Document summarisation

Sometimes the right answer is a summary. Recursive summarisation handles documents bigger than the context window.

Summarisation

Demos & docs

Build it. Read it. Try it.

Working console demos on GitHub, step-by-step how-to guides on the docs site, and the API reference for the classes used on this page.

Demo

Install the SDK

Ready to build document intelligence?

The most advanced local document Q&A technology. Semantic RAG and layout analysis. 100% on your infrastructure.

Download free View pricing

Ask your documents. Get grounded answers.

Semantic RAG meets document layout analysis.

Document import

Auto-detection

Extraction

Semantic RAG

Content-driven selection

Fast, direct processing

VLM-powered analysis

Multi-document queries

Source attribution

Intelligent caching

Smart context

Multi-turn dialogue

100% local

Agentic capabilities

Chat with PDF demo

Contract analysis

Financial review

Technical documentation

Research & academia

Compliance & audit

Customer support

Choose your models

PdfChat

DocumentRag

IVectorStore

VlmOcr

OCR

Document to Markdown

Document RAG engine

Document summarisation

Chat with PDF

Chat with PDF documents

Build a private document Q&amp;A

PdfChat

`PdfChat`

`DocumentRag`

`IVectorStore`

`VlmOcr`

Build a private document Q&A