Table of Contents

What You Can Build with LM-Kit.NET

Standalone .NET samples across multiple capability areas. Every sample runs entirely on your machine with no cloud API calls, no internet dependency, and no data leaving your infrastructure. Clone the repo, open a project, and run.

Browse all samples on GitHub →

Tip

New to LM-Kit.NET? Start with Your First AI Agent to set up your environment and build a working agent in minutes.

1. Get Running in 60 Seconds

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

Open any .csproj in your IDE and run. Models download automatically on first launch. No API keys. No configuration files. No cloud accounts.

Platform GPU Support
Windows x64 CUDA 12/13, Vulkan, CPU
Linux x64 / ARM64 CUDA 12/13, Vulkan, CPU
macOS (Apple Silicon) Metal, CPU

2. Solve a Real Problem

Find the right sample for your use case:

I need to... Start with
Have a conversation with a local LLM Multi-Turn Chat
Build an autonomous agent that reasons and acts Research Assistant
Ask questions about my documents Conversational RAG
Chat with a specific PDF Chat with PDF
Give an agent web search, calculators, and file access Tool Calling Assistant
Extract structured data from unstructured text Structured Data Extraction
Parse resumes (text, PDF, or scanned images) into structured profiles Resume Parser
Extract expense data from receipts (text, PDF, or photos) Receipt & Expense Scanner
Extract key terms and clauses from contracts Contract Key Terms Extractor
Detect and redact PII from documents PII Extraction
Classify documents by category Document Classification
Analyze customer sentiment at scale Sentiment Analysis
Transcribe audio locally Speech to Text
Convert scanned documents to searchable Markdown Document to Markdown
Extract text and coordinates from images with OCR VLM OCR with Coordinates
Build a help desk knowledge base Help Desk Knowledge Base
Translate content between languages Text Translator
Fine-tune a model on my own data Fine-Tuning
Turn meeting recordings into action items Smart Meeting Assistant
Analyze CSV or JSON data locally Data Analyst Agent
Classify and draft responses to emails Email Triage Agent
Monitor agent performance with telemetry Telemetry & Observability

3. Industry Solutions

Healthcare & Life Sciences

Patient data cannot leave your network. Local AI keeps PHI on-premises.

Capability Sample
Redact patient PII before sharing records PII Extraction, Batch PII Extraction
Transcribe clinical dictations locally Speech to Text
Generate meeting notes with action items Smart Meeting Assistant
Extract structured data from lab reports Structured Data Extraction
Classify medical documents by type Document Classification

Legal & Compliance

Attorney-client privilege requires that documents stay within your infrastructure.

Capability Sample
Review contracts and flag compliance issues Multi-Agent Document Review
Extract named entities from legal filings Named Entity Recognition
Search case files with natural language Conversational RAG
Summarize lengthy documents Document Summarizer
Extract key terms and risk flags from contracts Contract Key Terms Extractor

Also see the how-to guide: Process Email Archives for Compliance and Legal Discovery

Finance & Accounting

Regulatory requirements demand auditability and data sovereignty.

Capability Sample
Extract invoice data from PDFs and images Invoice Data Extraction
Scan receipts and extract expense data Receipt & Expense Scanner
Analyze market sentiment from reports Sentiment Analysis
Classify financial documents by type Batch Document Classification
Analyze financial data locally with AI Data Analyst Agent
Detect emotions in customer communications Emotion Detection

Manufacturing & Field Operations

Field devices and factory floors often have no internet access.

Capability Sample
Transcribe inspection notes offline Speech to Text
Extract text from equipment manuals (OCR) VLM OCR
Build a searchable knowledge base for procedures Help Desk Knowledge Base
Detect language in multilingual documents Language Detection

Also see the how-to guide: Build and Deploy an Offline AI Application for Edge Environments

Customer Service

Respond faster with AI that understands your product documentation.

Capability Sample
Build a support chatbot grounded in your docs Conversational RAG
Route tickets to the right team Custom Classification
Detect sarcasm and escalation cues Sarcasm Detection
Triage and draft email responses Email Triage Agent
Extract structured info from web content Web Content Extractor to JSON

4. Architecture Patterns

Each sample demonstrates a proven pattern you can adapt to your application:

Single Agent

One model, one task. The agent reasons, calls tools, and returns a result.

→ Research Assistant · Tool Calling Assistant · Web Search Assistant

Multi-Agent Pipeline

Multiple agents execute in sequence, each transforming the output for the next stage.

→ Content Creation Pipeline · Multi-Agent Document Review · Smart Meeting Assistant

Supervisor Delegation

A supervisor agent routes tasks to specialized worker agents based on the request.

→ Smart Task Router · Email Triage Agent

RAG (Retrieval-Augmented Generation)

Embed documents into vectors, retrieve relevant chunks, and generate grounded answers.

→ Conversational RAG · Single-Turn RAG · Help Desk Knowledge Base · Qdrant Integration

Document Processing Pipeline

Ingest documents, extract text and structure, convert formats, and index for search.

→ Document to Markdown · VLM OCR · Document Splitting · Document Processing Agent

External Tool Servers (MCP)

Connect agents to external tool servers via the Model Context Protocol.

→ MCP Integration · MCP Stdio Integration

5. Learning Paths

Beginner: Your First Week

Build confidence with core capabilities.

  1. Single-Turn Chat: send one prompt, get one response.
  2. Multi-Turn Chat: hold a conversation with history.
  3. Sentiment Analysis: classify text by sentiment.
  4. Text Translator: translate between languages.
  5. Speech to Text: transcribe audio locally.

Intermediate: Production Capabilities

Combine multiple features into real applications.

  1. Tool Calling Assistant: give the model tools to call.
  2. Conversational RAG: ground answers in your documents.
  3. PII Extraction: detect and redact sensitive data.
  4. Document to Markdown: convert any document to searchable text.
  5. Persistent Memory Assistant: remember facts across sessions.

Advanced: Enterprise Patterns

Build multi-agent systems with orchestration, delegation, and observability.

  1. Research Assistant: ReAct planning with web search and tool use.
  2. Smart Task Router: supervisor-based delegation across specialist agents.
  3. Content Creation Pipeline: sequential multi-agent workflows.
  4. Filter Pipeline: add middleware guardrails to agent conversations.
  5. Telemetry & Observability: instrument agents with OpenTelemetry.

6. All Samples at a Glance

AI Agents

Sample What It Demonstrates
Content Creation Pipeline Sequential multi-agent workflow
Data Analyst Agent Local data analysis with built-in tools
Document Processing Agent PDF and image processing with built-in tools
Email Triage Agent Email classification and response drafting
Filter Pipeline Middleware guardrails for agent conversations
MCP Integration Connect to external MCP tool servers
MCP Stdio Integration Stdio-based MCP server connectivity
News Monitoring Agent Automated news tracking with web search
Multi-Agent Document Review Parallel multi-perspective document analysis
Multi-Turn Chat with Memory Agent memory with semantic recall
Persistent Memory Assistant Long-term knowledge across sessions
Research Assistant ReAct planning with web search
Skill-Based Assistant Load agent skills from SKILL.md files
Smart Meeting Assistant Audio transcription to meeting notes pipeline
Smart Task Router Supervisor-based task delegation
Tool Calling Assistant Custom tool implementation
Web Search Assistant Real-time web search integration

Chat & Conversation

Sample What It Demonstrates
Chat Playground (MAUI) Cross-platform desktop chat app
Multi-Turn Chat Basic multi-turn conversation
Multi-Turn Chat with Chat History Guidance History management strategies
Multi-Turn Chat with Coding Assistant Code generation and review
Multi-Turn Chat with Custom Sampling Temperature, top-p, and sampler control
Multi-Turn Chat with MCP MCP tool integration in chat
Multi-Turn Chat with Persistent Session Save and restore conversation state
Multi-Turn Chat with Tools Function calling in multi-turn chat
Multi-Turn Chat with Vision Image understanding in conversations
Multi-Turn Chat with Yes/No Assistant Constrained output formatting
Single-Turn Chat Simplest possible interaction

Classification & Analysis

Sample What It Demonstrates
Batch Document Classification Classify documents at scale
Custom Classification Define your own categories
Document Classification Automatic document type detection
Emotion Detection Identify emotions in text
Keyword Extraction Pull key terms from documents
Language Detection from Document Detect language from PDFs and images
Sarcasm Detection Detect sarcasm and irony
Sentiment Analysis Positive/negative/neutral classification

Data Extraction

Sample What It Demonstrates
Batch PII Extraction PII detection across many documents
Contract Key Terms Extractor Extract clauses, parties, dates, and risk flags from contracts
Invoice Data Extraction Extract fields from invoices
Named Entity Recognition Identify people, places, organizations
PII Extraction Detect personal information
Receipt & Expense Scanner Extract expense data from receipts and photos using VLMs
Resume Parser Extract candidate profiles from resumes and scanned images using VLMs
Structured Data Extraction Schema-driven field extraction
Web Content Extractor to JSON Parse web content into structured JSON

Document Processing

Sample What It Demonstrates
Chat with PDF Interactive PDF Q&A
Document Processing Agent Agent-driven PDF and image processing
Document Splitting Intelligent document segmentation
Document Summarizer Automatic document summarization
Document to Markdown Convert any document to Markdown
Image to Markdown OCR images to Markdown
Text Summarizer Summarize plain text
VLM OCR Vision-based OCR
VLM OCR with Coordinates OCR with bounding box positions

Embeddings & RAG

Sample What It Demonstrates
Conversational RAG Multi-turn retrieval-augmented generation
Help Desk Knowledge Base Build a searchable support system
Image Similarity Search Find visually similar images
Retrieval Quality Tuning Optimize retrieval precision and recall
Single-Turn RAG Simple question-answering over documents
Single-Turn RAG with Qdrant External vector database integration

Integrations (Microsoft AI Ecosystem)

Sample What It Demonstrates
Microsoft.Extensions.AI Integration Use LM-Kit as an IChatClient drop-in
Semantic Kernel Integration Memory Demo LM-Kit as a Semantic Kernel backend

Also see the how-to guide: Migrate from Cloud AI APIs to Local Inference with Microsoft.Extensions.AI

Function Calling

Sample What It Demonstrates
Function Calling Register and invoke custom functions

Model Operations

Sample What It Demonstrates
Fine-Tuning Train a model on custom data (LoRA)
Quantization Reduce model size for edge deployment

Observability & Monitoring

Sample What It Demonstrates
Telemetry & Observability OpenTelemetry tracing and metrics

Prompt Templates

Sample What It Demonstrates
Prompt Templates with Logic Variables, conditionals, loops, and filters

Speech & Audio

Sample What It Demonstrates
Audio Transcription App (MAUI) Cross-platform transcription desktop app
Smart Meeting Assistant Transcription to meeting notes pipeline
Speech to Text Local Whisper transcription

Also see the how-to guide: Build a Voice-Commanded Agent That Executes Tools

Text Generation & Transformation

Sample What It Demonstrates
Text Corrector Grammar and spelling correction
Text Rewriter Rewrite in different styles
Text Translator Multilingual translation

7. Hardware Quick Reference

All samples run on CPU. GPU acceleration is optional but recommended for larger models.

Sample Category Minimum VRAM Recommended Model
Chat & Conversation 2 GB gemma3:4b
AI Agents (with tools) 4 GB qwen3:8b
Classification & Analysis 2 GB gemma3:4b
Data Extraction (NER, PII) 2 GB gemma3:4b
Document Processing (OCR) 4 GB gemma3:4b + vision model
Embeddings & RAG 2 GB qwen3-embedding:0.6b + chat model
Speech & Audio < 1 GB whisper-large-turbo3
Fine-Tuning 8+ GB Depends on base model
Multi-Agent Orchestration 4 GB qwen3:8b

No GPU? Every sample works on CPU. Use smaller models (gemma3:1b, qwen3:0.6b, phi4-mini:3.8b) for faster CPU inference. See Choosing the Right Model.

8. Platform & Integration Support

LM-Kit.NET fits into your existing .NET stack:

Integration What It Provides Sample
Microsoft.Extensions.AI IChatClient and IEmbeddingGenerator interfaces Extensions.AI Integration
Semantic Kernel Use LM-Kit as a Semantic Kernel AI backend Semantic Kernel Memory Demo
Model Context Protocol (MCP) Connect to external tool servers MCP Integration
Qdrant External vector database for production RAG Single-Turn RAG with Qdrant
OpenTelemetry Distributed tracing and metrics Telemetry & Observability
.NET MAUI Cross-platform desktop and mobile apps Chat Playground

9. What's New

Recently added to the how-to guides:

Explore all how-to guides: How-To Guides

10. Share & Get Involved

Found a sample useful? Use the share buttons below this article (copy link, X, LinkedIn, Reddit) to share it with your team.

⭐ Star the samples repo on GitHub · Report an issue · Contribute a sample

Share