Crate llm_transpile 

llm-transpiler

A high-performance Rust library that converts raw documents (Markdown, HTML, Plain Text, Tables, etc.) into a structured bridge format so LLM agents can receive maximum information with minimum tokens.

Quick Start

use llm_transpile::{transpile, FidelityLevel, InputFormat};

let md = "# Contract\n\nThis agreement was concluded in 2024.";
let result = transpile(md, InputFormat::Markdown, FidelityLevel::Semantic, Some(4096))
    .expect("transpile failed");
println!("{}", result);

Streaming Usage

use llm_transpile::{transpile_stream, FidelityLevel, InputFormat};
use futures::StreamExt;

async fn example() {
    let md = "# Document\n\nThis is a paragraph.";
    let mut stream = transpile_stream(md, InputFormat::Markdown, FidelityLevel::Semantic, 4096).await;
    while let Some(chunk) = stream.next().await {
        let chunk = chunk.expect("stream error");
        print!("{}", chunk.content);
        if chunk.is_final { break; }
    }
}

Structs

AdaptiveCompressor

Budget-based adaptive document compressor.

CompressionConfig

Context provided when running the compressor.

IRDocument

The complete IR representation of a parsed document.

StreamingTranspiler

Tokio channel-based streaming transpiler.

SymbolDict

Bidirectional mapping table between technical terms and PUA symbols.

TranspileChunk

A single output unit produced by the streaming transpiler.

Enums

CompressionStage

Compression stage enumeration.

DocNode

A semantic unit that makes up a document.

FidelityLevel

The degree of information loss permitted during document conversion.

InputFormat

Input document format.

StreamError

Streaming transpile error.

TranspileError

Transpile error.

Constants

MAX_INPUT_BYTES

Maximum input size accepted by transpile and transpile_stream. Inputs larger than this limit are rejected with TranspileError::InputTooLarge to prevent resource exhaustion on unbounded documents.

Functions

build_yaml_header

Builds a YAML header block from the IRDocument’s metadata.

linearize_table

Converts a table into token-efficient text.

render_full

Renders an entire IRDocument as a bridge-format string.

render_node

Renders a single DocNode as bridge text.

token_count

Returns the approximate token count for the given text.

transpile

Converts a document synchronously into the bridge format.

transpile_stream

Converts a document into a Tokio stream.