Named Entity Recognition for .NET, Multimodal NER with Custom Types, LM-Kit

Problem

Unstructured data is everywhere.

Contracts, emails, documents, and web content contain critical information buried in free-form text. Manual extraction is slow, error-prone, and impossible to scale.

Traditional approaches fall short

Regex patterns break with format variations
Rule-based systems require constant maintenance
Cloud APIs expose sensitive documents to third parties
Fixed entity types cannot adapt to domain-specific needs
Images and scanned documents require separate pipelines

LM-Kit NER delivers

LLM-powered extraction understands context and meaning
Zero-maintenance with built-in intelligence
100% on-device processing protects your data
Define unlimited custom entity types for your domain
Unified API for text and images with built-in OCR

Capabilities

Enterprise-grade entity extraction.

Extract structured information from any content with precision, flexibility, and complete data privacy.

Multimodal

Multimodal processing

Extract entities from text, images, and documents with a unified API. Built-in OCR handles scanned documents, screenshots, and photos automatically.

Custom

Custom entity types

Beyond built-in types, define unlimited custom entities with descriptions. Extract product codes, patient IDs, invoice numbers, or any domain-specific identifier.

Tracking

Position tracking

Every extracted entity includes precise character offsets for all occurrences. Enable highlighting, redaction, and targeted document manipulation.

Built-in types

8 standard entity types out of the box.

Start extracting immediately with production-ready entity definitions. Add your own custom types as needed.

Type

Person

Names of individuals, executives, authors, contacts

Type

Organization

Companies, institutions, agencies, government bodies

Type

Location

Places, addresses, cities, countries, regions

Type

Date

Calendar dates, time references, deadlines

Type

Money

Currency amounts, financial values, prices

Type

Percent

Percentage values, ratios, growth rates

Type

Product

Product names, brands, model numbers

Type

Custom

Define any entity type for your domain

Quick start

Extract entities in 5 lines of code.

The NamedEntityRecognition class provides a simple, intuitive API for extracting entities from any content. Load a model, create the recognizer, and call Recognize(). All built-in entity types are enabled by default.

Works with any compatible LM-Kit model
Synchronous and asynchronous methods available
Returns entity value, type, and all occurrences
Confidence scores for quality assessment
Multilingual support out of the box

API reference View demo

BasicNER.cs

using LMKit.Model;
using LMKit.TextAnalysis;

// Load the language model
var model = LM.LoadFromModelID("lmkit-tasks:4b-preview");

// Create the NER engine with default entity types
var ner = new NamedEntityRecognition(model);

// Extract entities from text
string text = "Apple Inc. announced that CEO Tim Cook " +
    "will visit Paris on January 15, 2025.";

var entities = ner.Recognize(text);

foreach (var entity in entities)
{
    Console.WriteLine($"{entity.Type}: {entity.Value}");
}

Customization

Define your own entity types.

Extract domain-specific information by defining custom entity types with EntityDefinition. Provide a name and optional description to guide extraction. Mix built-in and custom types freely.

Define unlimited custom entity types
Optional descriptions improve extraction accuracy
Mix with built-in types as needed
Perfect for product codes, IDs, domain terms
Use Guidance property for complex extraction rules

CustomEntities.cs

var model = LM.LoadFromModelID("lmkit-tasks:4b-preview");

// Define custom entity types with descriptions
var definitions = new List<EntityDefinition>
{
    new EntityDefinition(NamedEntityType.Person),
    new EntityDefinition(NamedEntityType.Organization),
    new EntityDefinition(NamedEntityType.Custom,
        "ProductCode",
        "Product codes like SKU-12345 or ITEM-ABC"),
    new EntityDefinition(NamedEntityType.Custom,
        "PatientID",
        "Medical record numbers")
};

var ner = new NamedEntityRecognition(model);
foreach (var def in definitions) ner.EntityDefinitions.Add(def);

Multimodal

Extract entities from images.

Process scanned documents, screenshots, business cards, and photos with the same API. Built-in OCR extracts text automatically, then NER identifies entities. Use OcrEngine for traditional OCR or rely on vision language models for intelligent extraction.

Same API for text and image inputs
Built-in OCR integration via OcrEngine property
Vision language model support for complex layouts
PreferredInferenceModality for control over processing
Process PDFs, JPEGs, PNGs, and more

NerFromImage.cs

using LMKit.Model;
using LMKit.TextAnalysis;
using LMKit.Data;

// Load a vision-capable model
var model = LM.LoadFromModelID("qwen3-vl:8b");
var ner = new NamedEntityRecognition(model)
{
    // Control inference modality
    PreferredInferenceModality = InferenceModality.Multimodal
};

// Extract entities from a scanned business card
var attachment = new Attachment("business_card.jpg");
var entities = ner.Recognize(attachment);

Position tracking

Track every entity mention.

Each extracted entity includes an Occurrences collection with precise character offsets for every mention in the source text. Enable precise highlighting, redaction, entity linking, and document manipulation workflows.

Start and end character positions for each occurrence
Multiple mentions tracked for the same entity
Enable precise document redaction
Build entity highlighting and annotation features
Power entity linking and knowledge graph construction

TrackOccurrences.cs

var model = LM.LoadFromModelID("lmkit-tasks:4b-preview");
var ner = new NamedEntityRecognition(model);

string text = "Microsoft CEO Satya Nadella announced " +
    "that Microsoft will invest in AI. " +
    "Nadella emphasized Microsoft's commitment.";

var entities = ner.Recognize(text);

foreach (var entity in entities)
{
    Console.WriteLine($"{entity.Type}: {entity.Value}");
    foreach (var occ in entity.Occurrences)
        Console.WriteLine($"  [{occ.Start}-{occ.End}]");
}

Workflow

Four steps to structured data.

From raw content to structured entities in milliseconds, all running locally on your hardware.

Step 1

Load model

Choose from LM-Kit's optimized task models or bring your own. Models run entirely on-device.

Step 2

Configure entities

Use default entity types or define custom ones for your domain. Add guidance for complex extraction.

Step 3

Process content

Feed text or images through the recognizer. Built-in OCR handles visual content automatically.

Step 4

Get results

Receive structured entities with types, values, confidence scores, and position offsets.

Applications

Real-world use cases.

Organizations across industries use LM-Kit NER to automate document processing, enhance search, and ensure compliance.

Legal

Contract analysis

Extract parties, dates, amounts, and obligations from legal documents. Automate contract review and clause identification.

Finance

Invoice processing

Parse vendor names, amounts, dates, and line items from invoices and receipts. Enable touchless AP automation.

Compliance

Compliance monitoring

Identify regulated entities in communications. Monitor for mentions of competitors, products, or restricted topics.

Health

Medical records

Extract patient names, medications, diagnoses, and providers from clinical notes. Enable structured EHR data entry.

Search

Enhanced search

Index documents by extracted entities. Enable faceted search by person, organization, location, or date.

News

News intelligence

Track mentions of companies, executives, and competitors across news feeds. Build real-time market intelligence.

Graphs

Knowledge graphs

Extract entities and relationships to build knowledge graphs. Connect people, organizations, and events automatically.

Support

Support ticket routing

Extract products, issues, and customer details from support tickets. Enable intelligent routing and prioritization.

HR

Resume parsing

Extract candidate names, employers, skills, and education from resumes. Automate applicant tracking workflows.

Comparison

LM-Kit vs. alternatives.

See how LM-Kit NER compares to cloud APIs and traditional NLP libraries.

Feature	LM-Kit NER	Cloud NER APIs	Traditional NLP
Data privacy	100% on-device	Data sent to cloud	Local processing
Custom entity types	Unlimited	Limited or none	Requires retraining
Multimodal (images)	Built-in	Separate service	Not supported
Position tracking	Yes	Varies	Yes
Contextual understanding	LLM-powered	LLM-powered	Rule-based
Multilingual	Yes	Yes	Per-language models
Offline operation	Yes	No	Yes
Per-request cost	None	Per API call	None

Developer Resources

API reference.

Complete documentation for the NamedEntityRecognition class and related types.

`NamedEntityRecognition`

Main class for extracting named entities from text and images with configurable entity types.

View docs

`EntityDefinition`

Define built-in or custom entity types with optional descriptions for guided extraction.

View docs

`NamedEntityType`

Enumeration of built-in entity types: Person, Organization, Location, Date, Money, Percent, Product, Event, Custom.

View docs

`Recognize`

Synchronously extract entities from text or image attachments with optional cancellation.

View docs

`RecognizeAsync`

Asynchronously extract entities for non-blocking operation in UI and server applications.

View docs

`Guidance`

Property to provide semantic guidance for domain-specific extraction rules and context.

View docs

`Confidence`

Get the confidence score of the last recognition operation for quality assessment.

View docs

`OcrEngine`

Optional OCR engine for traditional text extraction from raster content during recognition.

View docs

`MaximumContextLength`

Configure maximum context length in tokens for processing large documents.

View docs

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

Ready to extract structured data?

8+ entity types. Custom definitions. Multimodal. Position tracking. 100% on-device. Start building intelligent .NET applications today.

Download free API documentation

Extract structured data from any content.

Traditional approaches fall short

LM-Kit NER delivers

Multimodal processing

Custom entity types

Position tracking

Person

Organization

Location

Date

Money

Percent

Product

Custom

Load model

Configure entities

Process content

Get results

Contract analysis

Invoice processing

Compliance monitoring

Medical records

Enhanced search

News intelligence

Knowledge graphs

Support ticket routing

Resume parsing

NamedEntityRecognition

EntityDefinition

NamedEntityType

Recognize

RecognizeAsync

Guidance

Confidence

OcrEngine

MaximumContextLength

Named entity recognition

Extract named entities

Validate extracted entities

NamedEntityRecognition

`NamedEntityRecognition`

`EntityDefinition`

`NamedEntityType`

`Recognize`

`RecognizeAsync`

`Guidance`

`Confidence`

`OcrEngine`

`MaximumContextLength`