Probe on GitHub

Project Overview: Probe was born from a simple observation: we read code 10x more than we write it, and AI agents read even more than that. Yet most AI coding tools still explore codebases with naive grep-style text search, or with embedding pipelines that require heavy indexing and return probabilistic, fragmented matches. Both approaches fall apart on real, enterprise-scale codebases.

Probe treats code as structured data instead of plain text. It combines ripgrep-level scanning speed with tree-sitter AST parsing, so a search returns complete functions, classes, and semantic blocks — not arbitrary text fragments. Queries use familiar Elasticsearch-style boolean syntax with BM25 ranking, and results respect a configurable token budget so they fit cleanly into an LLM context window.

The part I care most about is the zero-setup philosophy: no indexing step, no embedding models, no external services. Point it at a multi-million line repository and it just works — deterministically, on your machine. Written in Rust for performance, with a Node.js SDK and a built-in agent mode, it has become the code intelligence foundation for AI assistants and the first project of Probe Labs, the home I created for my AI developer-tools work.

Outcome

Probe has grown into a widely used building block for AI coding workflows, with nearly 700 GitHub stars and adoption in tools that need reliable code context at enterprise scale. It also became the foundation the rest of the Probe Labs ecosystem — including Visor and Proof — is built on.