semble_rs
A Rust command-line tool that replaces grep, cat and ls for AI coding agents, returning hybrid BM25 and embedding code search, a compact codebase tree, dependency and impact analysis, and compressed build or CI logs.
semble_rs is a Rust port and superset of the Python project MinishLab/semble, built to cut the tokens an AI coding agent spends reading a codebase. It combines BM25 keyword search with Model2Vec static embeddings, code-aware reranking and AST-based chunking via Tree-sitter to return the exact code chunks a query needs, and prints a token-cheap tree of a project instead of a full directory listing.
It also builds a dependency graph to show what a file imports and what breaks if it changes, and its digest command compresses build, test and CI output, for example cutting a 3 MB CI log to 35 KB. It ships as a single binary with no daemon, API keys or GPU requirement, and integrates with agents such as Claude Code, Codex, Cursor and Aider.