Lightweight Stream Tokenizers for syntax highlighting and formatting. Perfect building block for render logic and handling AI responses.
NTokenizers sits in the middle of a tokenization pipeline — it takes raw source code or markup as a stream and emits a sequence of typed tokens that a downstream renderer can consume:
┌────────┐ stream ┌──────────────┐ tokens ┌────────────┐
│ source │ ────────────► │ NTokenizers │ ──────────► │ Renderer │ ──► styled output
└────────┘ └──────────────┘ └────────────┘
This separation of concerns means NTokenizers stays format-focused while rendering is delegated to the consumer — whether that is a console UI, a web component, or a custom formatter. The stream-first design ensures low memory usage and real-time compatibility with AI chat outputs, CI logs, or any scenario where data arrives incrementally.
- NTokenizers.Extensions.Spectre.Console Spectre.Console rendering extensions for NTokenizers, Style-rich console syntax highlighting.
NTokenizers provides a collection of stream-capable tokenizers for processing structured text. Each tokenizer breaks down input into meaningful tokens as data arrives in real-time—ideal for large files or streaming data without loading everything into memory.
The library supports the following formats:
- Markup languages: Markdown, HTML
- Data formats: JSON, YAML, TOML, XML
- Programming languages: C#, C, C++, Go, Java, Kotlin, Python, Rust, SQL, Swift, TypeScript, CSS
The MarkdownTokenizer acts as a composite tokenizer, delegating code blocks to the appropriate sub-tokenizer based on the language tag. This allows seamless parsing of documents that mix multiple formats in a single pass.
Initialize any tokenizer and start parsing a stream:
// Use any tokenizer — replace [Language] with the target format
await [Language]Tokenizer.Create().ParseAsync(stream, onToken: async token =>
{
// Handle tokens as they arrive
});NTokenizers is a .NET library written in C# that provides tokenizers for processing structured text formats like Markdown, JSON, XML, HTML, YAML, TOML, SQL, Typescript, CSS, CSharp, C, C++, Go, Java, Kotlin, Rust, Swift and Python. The Tokenize method is the core functionality that breaks down structured text into meaningful components (tokens) for processing. Its key feature is stream processing capability - it can handle data as it arrives in real-time, making it ideal for processing large files or streaming data without loading everything into memory at once.
Warning
These tokenizers are not validation-based and are primarily intended for prettifying, formatting, or visualizing structured text. They do not perform strict validation of the input format, so they may produce unexpected results when processing malformed or invalid XML, JSON, or HTML. Use them with caution when dealing with untrusted or poorly formatted input.
Most tokenizers, such as json, xml, or etc..., can be used individually, depending on the specific format you want to parse.
The MarkdownTokenizer however is a special case. Instead of working on a single format, it acts as a composite tokenizer, using the other tokenizers as subtokenizers. When parsing a stream, MarkdownTokenizer delegates portions of the input to the appropriate subtokenizer, allowing it to handle multiple formats seamlessly in one pass.
The same principle applies to inline tokenizers such as Heading, Blockquote, ListItem, and others. However, they cannot be used individually and produce the same token types as the MarkdownTokenizer.
┌─────────┐
│ stream │
└─────────┘
│ ParseAsync()
▼
┌─────────────────────┐
│ MarkdownTokenizer │ ───────────► fire markdown tokens
└─────────────────────┘
│
▼ ┌─────────┐
├──────►│ json │ ───► fire json tokens
│ └─────────┘
│
│ ┌─────────┐
├──────►│ Heading │ ───► fire markdown tokens
│ └─────────┘
│
│ ┌─────────┐
├──────►│ html │ ───► fire html tokens
│ └─────────┘
│ │
│ ▼ ┌─────────┐
│ ├──────►│ css │ ───► fire css tokens
│ │ └─────────┘
│ │
│ │ ┌─────────┐
│ └──────►│ script │ ───► fire typescript tokens
│ └─────────┘
│ ┌─────────┐
└──────►│ etc.. │ ───► etc
└─────────┘
Here's a simple example showing how to use the MarkdownTokenizer:
using NTokenizers.Core;
using NTokenizers.Css;
using NTokenizers.Html;
using NTokenizers.Json;
using NTokenizers.Markdown;
using NTokenizers.Markdown.Metadata;
using NTokenizers.Typescript;
using NTokenizers.Xml;
using Spectre.Console;
using System.Diagnostics;
using System.IO.Pipes;
using System.Text;
class Program
{
static async Task Main()
{
string markdown = """
Here is some **bold** text and some *italic* text.
# NTokenizers Showcase
## Css example
```css
.user {
color: #FFFFFF;
active: true;
}
```
## XML example
```xml
<user id="4821" active="true">
<name>Laura Smith</name>
</user>
```
## HTML example
```html
<html>
<head>
<style>
body { font-family: Arial, sans-serif; background-color: #f0f8ff; }
.header { color: #4682b4; text-align: center; }
.content { margin: 20px; padding: 15px; background-color: white; border-radius: 5px; }
</style>
</head>
<body>
<p>Hello world!</p>
<script>
console.log("Hello from the sample script!");
document.addEventListener('DOMContentLoaded', function() {
console.log("DOM is fully loaded");
});
</script>
</body>
</html>
```
## JSON example
```json
{
"name": "Laura Smith",
"active": true
}
```
## TypeScript example
```typescript
const user = {
name: "Laura Smith",
active: true
};
```
""";
// Create connected streams
using var pipe = new AnonymousPipeServerStream(PipeDirection.Out);
using var reader = new AnonymousPipeClientStream(PipeDirection.In, pipe.ClientSafePipeHandle);
// Start slow writer
var writerTask = EmitSlowlyAsync(markdown, pipe);
// Parse markup
await MarkdownTokenizer.Create().ParseAsync(reader, onToken: async token =>
{
if (token.Metadata is ICodeBlockMetadata codeBlock)
{
AnsiConsole.WriteLine();
AnsiConsole.Write(new Markup($"[bold lime]{codeBlock.Language}:[/]"));
AnsiConsole.WriteLine();
}
if (token.Metadata is ListItemMetadata listMetadata)
{
AnsiConsole.Write(new Markup($"[bold lime]{listMetadata.Marker} [/]"));
await listMetadata.RegisterInlineTokenHandler(inlineToken =>
{
var value = Markup.Escape(inlineToken.Value);
AnsiConsole.Write(new Markup($"[bold red]{value}[/]"));
});
Debug.WriteLine("Written listItem inlines");
}
else if (token.Metadata is HeadingMetadata headingMetadata)
{
await headingMetadata.RegisterInlineTokenHandler(inlineToken =>
{
var value = Markup.Escape(inlineToken.Value);
var colored = headingMetadata.Level != 1 ?
new Markup($"[bold GreenYellow]{value}[/]") :
new Markup($"[bold yellow]** {value} **[/]");
AnsiConsole.Write(colored);
});
Debug.WriteLine("Written Heading inlines");
}
else if (token.Metadata is XmlCodeBlockMetadata xmlMetadata)
{
await xmlMetadata.RegisterInlineTokenHandler(inlineToken =>
{
var value = Markup.Escape(inlineToken.Value);
var colored = inlineToken.TokenType switch
{
XmlTokenType.ElementName => new Markup($"[blue]{value}[/]"),
XmlTokenType.OpeningAngleBracket => new Markup($"[yellow]{value}[/]"),
XmlTokenType.ClosingAngleBracket => new Markup($"[yellow]{value}[/]"),
XmlTokenType.SelfClosingSlash => new Markup($"[yellow]{value}[/]"),
XmlTokenType.AttributeName => new Markup($"[cyan]{value}[/]"),
XmlTokenType.AttributeEquals => new Markup($"[yellow]{value}[/]"),
XmlTokenType.AttributeQuote => new Markup($"[grey]{value}[/]"),
XmlTokenType.AttributeValue => new Markup($"[green]{value}[/]"),
XmlTokenType.Text => new Markup($"[white]{value}[/]"),
XmlTokenType.Whitespace => new Markup($"[grey]{value}[/]"),
_ => new Markup(value)
};
AnsiConsole.Write(colored);
});
}
else if (token.Metadata is JsonCodeBlockMetadata jsonMetadata)
{
await jsonMetadata.RegisterInlineTokenHandler(inlineToken =>
{
var value = Markup.Escape(inlineToken.Value);
var colored = inlineToken.TokenType switch
{
JsonTokenType.StartObject => new Markup($"[yellow]{value}[/]"),
JsonTokenType.EndObject => new Markup($"[yellow]{value}[/]"),
JsonTokenType.StartArray => new Markup($"[yellow]{value}[/]"),
JsonTokenType.EndArray => new Markup($"[yellow]{value}[/]"),
JsonTokenType.PropertyName => new Markup($"[cyan]{value}[/]"),
JsonTokenType.StringValue => new Markup($"[green]{value}[/]"),
JsonTokenType.Number => new Markup($"[magenta]{value}[/]"),
JsonTokenType.True => new Markup($"[orange1]{value}[/]"),
JsonTokenType.False => new Markup($"[orange1]{value}[/]"),
JsonTokenType.Null => new Markup($"[grey]{value}[/]"),
JsonTokenType.Colon => new Markup($"[yellow]{value}[/]"),
JsonTokenType.Comma => new Markup($"[yellow]{value}[/]"),
JsonTokenType.Whitespace => new Markup($"[grey]{value}[/]"),
_ => new Markup(value)
};
AnsiConsole.Write(colored);
});
}
else if (token.Metadata is HtmlCodeBlockMetadata htmlMetadata)
{
await htmlMetadata.RegisterInlineTokenHandler(async inlineToken =>
{
if (inlineToken.Metadata is TypeScriptCodeBlockMetadata tsMeta)
{
await HandleScript(tsMeta);
}
else if (inlineToken.Metadata is CssCodeBlockMetadata cssMeta)
{
await HandleCss(cssMeta);
}
else
{
var value = Markup.Escape(inlineToken.Value);
var colored = inlineToken.TokenType switch
{
HtmlTokenType.OpeningAngleBracket => new Markup($"[yellow]{value}[/]"),
HtmlTokenType.ClosingAngleBracket => new Markup($"[yellow]{value}[/]"),
HtmlTokenType.SelfClosingSlash => new Markup($"[yellow]{value}[/]"),
HtmlTokenType.AttributeName => new Markup($"[cyan]{value}[/]"),
HtmlTokenType.AttributeEquals => new Markup($"[yellow]{value}[/]"),
HtmlTokenType.AttributeQuote => new Markup($"[grey]{value}[/]"),
HtmlTokenType.AttributeValue => new Markup($"[green]{value}[/]"),
HtmlTokenType.Text => new Markup($"[white]{value}[/]"),
HtmlTokenType.Comment => new Markup($"[grey]{value}[/]"),
HtmlTokenType.Whitespace => new Markup($"[grey]{value}[/]"),
_ => new Markup(value)
};
AnsiConsole.Write(colored);
}
});
}
else if (token.Metadata is TypeScriptCodeBlockMetadata tsMetadata)
{
await HandleScript(tsMetadata);
}
else if (token.Metadata is CssCodeBlockMetadata cssMetadata)
{
await HandleCss(cssMetadata);
}
else
{
// Handle regular markup tokens
var value = Markup.Escape(token.Value);
var colored = token.TokenType switch
{
MarkdownTokenType.Text => new Markup($"{value}"),
MarkdownTokenType.Bold => new Markup($"[bold]{value}[/]"),
MarkdownTokenType.Italic => new Markup($"[italic]{value}[/]"),
_ => new Markup(value)
};
AnsiConsole.Write(colored);
}
if (token.Metadata is InlineMetadata)
{
AnsiConsole.WriteLine();
}
});
await writerTask;
Console.WriteLine();
Console.WriteLine("Done.");
}
private static async Task HandleScript(TypeScriptCodeBlockMetadata tsMetadata)
{
await tsMetadata.RegisterInlineTokenHandler(inlineToken =>
{
var value = Markup.Escape(inlineToken.Value);
var colored = inlineToken.TokenType switch
{
TypescriptTokenType.Identifier => new Markup($"[cyan]{value}[/]"),
TypescriptTokenType.Keyword => new Markup($"[blue]{value}[/]"),
TypescriptTokenType.StringValue => new Markup($"[green]{value}[/]"),
TypescriptTokenType.Number => new Markup($"[magenta]{value}[/]"),
TypescriptTokenType.Operator => new Markup($"[yellow]{value}[/]"),
TypescriptTokenType.Comment => new Markup($"[grey]{value}[/]"),
TypescriptTokenType.Whitespace => new Markup($"[grey]{value}[/]"),
_ => new Markup(value)
};
AnsiConsole.Write(colored);
});
}
private static async Task HandleCss(CssCodeBlockMetadata cssMetadata)
{
await cssMetadata.RegisterInlineTokenHandler(inlineToken =>
{
var value = Markup.Escape(inlineToken.Value);
var colored = inlineToken.TokenType switch
{
CssTokenType.Identifier => new Markup($"[white]{value}[/]"),
CssTokenType.Number => new Markup($"[magenta]{value}[/]"),
CssTokenType.Operator => new Markup($"[yellow]{value}[/]"),
CssTokenType.Selector => new Markup($"[yellow]{value}[/]"),
CssTokenType.Comment => new Markup($"[green]{value}[/]"),
CssTokenType.Whitespace => new Markup($"[grey]{value}[/]"),
_ => new Markup(value)
};
AnsiConsole.Write(colored);
});
}
static async Task EmitSlowlyAsync(string markdown, Stream output)
{
var rng = new Random();
byte[] bytes = Encoding.UTF8.GetBytes(markdown);
foreach (var b in bytes)
{
await output.WriteAsync(new[] { b }.AsMemory(0, 1));
await output.FlushAsync();
await Task.Delay(rng.Next(0, 2));
}
output.Close(); // EOF
}
}For more information, check out the documentation here.