Expand description
Tree-sitter based code compression for LLM context packs.
This module provides AST-aware compression that extracts structurally significant parts of source files — function signatures, type definitions, imports, and optionally top-level control flow — while discarding implementation bodies.
The result is a dramatically smaller representation (typically 50–70 % fewer tokens) that still preserves the full API surface and architectural overview of the original source.
§Supported languages
Sephera embeds Tree-sitter grammars for: Rust, Python, TypeScript, JavaScript, Go, Java, C++, and C.
Files in unsupported languages fall back to uncompressed head-excerpt behaviour automatically.
§Example
use sephera_compression::{
CompressionMode, SupportedLanguage, compress_source,
};
let source = b"fn add(a: i32, b: i32) -> i32 {\n a + b\n}\n";
let result = compress_source(source, SupportedLanguage::Rust, CompressionMode::Signatures);
assert!(result.is_ok());
let output = result.unwrap();
assert!(output.content.contains("fn add("));
assert!(output.content.contains("{ … }"));Re-exports§
pub use parser::SupportedLanguage;pub use parser::new_parser;pub use parser::with_parser;pub use types::CompressedOutput;pub use types::CompressionMode;
Modules§
Functions§
- compress_
source - Compresses a source file using Tree-sitter AST extraction.
- extract_
compressed - Extracts a compressed representation from a fully parsed Tree-sitter tree.