'sem' Released to Help LLMs Understand Code

sem is a Rust-based CLI tool that analyzes differences at the entity level—functions, classes, structs, etc.—rather than simple line-by-line diffs.

Conventional Git diffs contain a lot of unnecessary context and noise when fed into an LLM, wasting many tokens, but sem uses tree-sitter to extract only the core structural changes. This raises the LLM's attention score and enables more accurate reasoning with fewer tokens.

Key features:

• sem diff: Entity-level diff analysis with word-level highlighting
• sem impact: Analyzes dependencies affected when a specific entity changes
• sem context: Provides optimized context tailored to the LLM's token budget
• sem blame/log: Tracks change history and evolution at the entity level

It supports various languages such as Rust, Python, TypeScript, Go, and Java, as well as various file formats including JSON, YAML, and Markdown.