probelabs/probe

What it solves

Probe addresses the difficulty of providing AI agents with accurate, high-quality context from enterprise-scale codebases. Unlike traditional tools that rely on simple text searches (grep) or embedding-based vector searches (which can split functions mid-block and require slow indexing), Probe provides structural, AST-aware context that treats code as code rather than plain text.

How it works

Probe uses Tree-sitter AST parsing to understand the actual structure of code. It implements an Elasticsearch-style boolean query language (supporting AND, OR, NOT, and phrases) and BM25/TF-IDF ranking for semantic search without requiring a pre-built index or embedding model. It can return complete semantic blocks (like entire functions or classes) and includes a built-in agent that can translate user intent into these precise structural queries.

Who it’s for

It is designed for developers and AI engineers building coding assistants, as well as teams performing spec-driven development, code reviews, and onboarding on large codebases.

Highlights

• Zero Setup: No indexing or embedding services required; works instantly and offline.
• AST-Aware: Returns complete functions and classes instead of arbitrary text chunks.
• Multi-Mode Access: Available as an MCP server for AI editors (like Claude Code), a CLI tool, and a Node.js SDK.
• LLM Script: A sandboxed JavaScript DSL that allows the agent to execute multi-step analysis pipelines.
• Broad Language Support: Supports a wide array of languages including Rust, Python, TypeScript, Go, C++, Java, and more.