👉 Try the demo: https://github.com/LM-Kit/lm-kit-net-samples/tree/main/console_net/agents/document_processing_agent

Document Processing Agent for C# .NET Applications

🎯 Purpose of the Demo

The Document Processing Agent demo shows how to build an AI agent that processes PDFs and images using 9 built-in Document tools from the LM-Kit.NET SDK. The agent can inspect, split, merge, and render PDFs, deskew, crop, and resize scanned images, extract text from multi-format documents, and run OCR on images in 34 languages, all driven by natural language instructions.

👥 Industry Target Audience

This demo is useful for developers and businesses involved in:

Mailroom and scanning automation: an agent that inspects bulk-scanned PDFs, splits them by page ranges, and extracts text from each segment.
Document preparation pipelines: deskew, crop, and resize scanned images before feeding them to vision models or OCR engines.
Legal and compliance: split combined filings into individual documents, merge related documents, and extract text for review.
Healthcare and insurance: process mixed-format claims with PDF splitting, text extraction, and OCR in a single conversational workflow.

🚀 Problem Solved

Building document processing pipelines traditionally requires writing custom code for each step: PDF parsing, image manipulation, OCR integration, and file assembly. This demo shows how an AI agent with built-in tools can orchestrate all these operations from natural language instructions, reducing a multi-script workflow to a single prompt.

💻 Sample Application Description

The Document Processing Agent demo is a console app that:

Lets you select a tool-calling model (or enter a custom model URI)
Downloads and loads the model with progress feedback
Creates an agent equipped with all 9 Document tools
Accepts natural language tasks in an interactive loop
Displays tool invocations in real time as the agent works
Shows results with execution statistics (tool calls, duration, inferences)

✨ Key Features

9 built-in Document tools: PdfInfo, PdfSplit, PdfMerge, PdfRender, ImageDeskew, ImageCrop, ImageResize, DocumentText, and Ocr.
Natural language control: describe what you need and the agent selects the right tools.
Tool call monitoring: see which tools the agent invokes and what arguments it passes.
Multi-step workflows: chain operations like "deskew, crop, then OCR" in a single prompt.
Interactive console loop: process multiple tasks in one session.

💻 Minimal Integration Snippet

using LMKit.Model;
using LMKit.Agents;
using LMKit.Agents.Tools.BuiltIn;

// Load a tool-calling model
using LM model = LM.LoadFromModelID("qwen3:8b");

// Create agent with document tools
var agent = Agent.CreateBuilder(model)
    .WithPersona("Document Processing Assistant")
    .WithTools(tools =>
    {
        tools.Register(BuiltInTools.PdfSplit);
        tools.Register(BuiltInTools.PdfMerge);
        tools.Register(BuiltInTools.PdfInfo);
        tools.Register(BuiltInTools.DocumentText);
        tools.Register(BuiltInTools.Ocr);
        tools.Register(BuiltInTools.ImageDeskew);
        tools.Register(BuiltInTools.ImageCrop);
        tools.Register(BuiltInTools.ImageResize);
        tools.Register(BuiltInTools.PdfRender);
    })
    .Build();

// Run a document task
var result = await agent.RunAsync(
    "Extract pages 1-3 from 'report.pdf' into 'summary.pdf', " +
    "then get the text from page 1.");

Console.WriteLine(result.Content);

🛠️ Getting Started

📋 Prerequisites

.NET 8.0 or later
A tool-calling capable model (~6.5 GB VRAM for Qwen 3 8B)

📥 Download the Project

.NET Console Demo

▶️ Running the Application

Clone the repository:

git clone https://github.com/LM-Kit/lm-kit-net-samples

Navigate to the project directory:

cd lm-kit-net-samples/console_net/agents/document_processing_agent

Build and run the application:

dotnet build
dotnet run

Follow the on-screen prompts to select a model and enter document processing tasks.

💡 Example Usage

Select a tool-calling model: choose Qwen 3 8B (recommended) or another tool-calling model.
Enter a document task: type a natural language instruction like "Split pages 1-5 from report.pdf into intro.pdf".
Watch tool invocations: the agent shows which tools it calls and their arguments.
Review results: the agent reports what it did, with execution statistics.
Continue processing: enter more tasks or type 'q' to quit.

Example Prompts

How many pages does 'contract.pdf' have?
Extract pages 1-3 from 'report.pdf' into 'summary.pdf'
Merge 'part1.pdf' and 'part2.pdf' into 'combined.pdf'
Deskew 'scan.png', crop its borders, then resize to 1200x1600
Run OCR on 'receipt.jpg' in French
Render page 5 of 'manual.pdf' as a PNG at 2x zoom
Extract text from 'quarterly_report.docx'

🔍 Notes on Key Types

Agent (LMKit.Agents): the AI agent that reasons about tasks and calls tools autonomously. Created via Agent.CreateBuilder(model).
BuiltInTools (LMKit.Agents.Tools.BuiltIn): static factory class providing access to all 65 built-in tools. The Document category includes 9 tools for PDF and image processing.
AgentExecutionResult (LMKit.Agents): result of agent execution with Content (final answer), IsSuccess, ToolCalls (list of tool invocations), and Duration.

🔧 Extend the Demo

Add BuiltInTools.Json and BuiltInTools.Csv to parse extracted text into structured data.
Combine with DocumentSplitting to detect document boundaries before agent-driven splitting.
Use BuiltInTools.FileSystem to let the agent browse folders and process all PDFs in a directory.
Add BuiltInTools.WebSearch for an agent that can look up document metadata online.
Integrate with a multi-agent workflow where one agent preprocesses images and another extracts data.

Table of Contents