Probe – AI‑aware code‑search and reasoning engine

One‑line summary – Probe is a local, AST‑aware code‑search engine that can be driven directly from the command line or through a built‑in LLM‑agent. It lets large‑language‑model assistants understand whole functions, classes or structs in enterprise‑size repositories without any indexing or embedding step.

---

What it does (as described in the README)

• Parses source files with Tree‑sitter to build an abstract syntax tree (AST) for each supported language.
• Semantic search using Elasticsearch‑style boolean queries (AND, OR, NOT, field filters, file‑type filters, etc.).
• Returns complete code blocks (functions, classes, structs) rather than line fragments.
• Provides AST pattern matching (probe query) so you can search for structural templates like fn $NAME($$$) -> Result<$RET>.
• Zero‑setup indexing – the engine scans the filesystem on‑the‑fly; no embedding model, vector DB, or external service is required.
• Deterministic ranking with BM25/TF‑IDF (and optional hybrid BERT reranking) and token‑budget controls (--max‑tokens).
• Built‑in agent (probe agent) that talks to any LLM provider (Anthropic, OpenAI, Google, Bedrock) and orchestrates multi‑step searches, extraction, and even code edits.
• MCP integration – the agent can be registered as an MCP server for editors like Claude Code, allowing the editor to call Probe tools directly.
• Node.js SDK and Vercel AI SDK bindings so developers can embed the search/agent in their own applications.
• Fully local – all search work happens on your machine; only the LLM calls go out to the chosen provider.

---

Core features (concise list)

Feature	Detail
AST‑aware parsing	Uses Tree‑sitter for >20 languages (Rust, Python, TS/JS, Go, C/C++, Java, etc.).
Semantic search	Boolean query language, file‑type (ext:) and directory (dir:) filters, BM25/TF‑IDF ranking.
Zero indexing	No pre‑computed embeddings; instant results on any repo.
Deterministic output	Same query → same results, no stochastic vector similarity drift.
Token‑aware budgeting	--max‑tokens limits the amount of code sent to the LLM.
Built‑in LLM agent	Handles multi‑step reasoning, code editing, delegation, bash execution.
MCP server mode	Register as probe server for AI‑enabled editors (Claude Code, Codex, etc.).
CLI & SDK	`probe search
Hybrid ranking	Optional BERT reranker for higher‑quality results.
Cross‑platform install	npm global, curl script, PowerShell script, or build from source (Rust/Cargo).

---

Typical use‑cases

• Spec‑driven development – ask the agent “How is authentication implemented?” and get the full function(s) that implement it.
• Code reviews & onboarding – quickly surface all relevant symbols around a change without manually grepping.
• Automated refactoring – combine agent with --allow-edit to let the LLM propose and apply edits.
• Tooling integration – plug the CLI tools into custom IDE extensions or CI pipelines for static analysis.
• Offline environments – run on air‑gapped machines because search never requires external services.

---

Quick‑start (from the README)

# Install globally via npm (recommended)
npm install -g @probelabs/probe

# One‑shot question using the built‑in agent (any LLM provider)
npx -y @probelabs/probe agent "How is authentication implemented?"

# Semantic search from the terminal
npx -y @probelabs/probe search "authentication AND login" ./src

# AST pattern query (Rust example)
npx -y @probelabs/probe query "fn $NAME($$$) -> Result<$RET>" --language rust

For editor integration, add a small JSON block to ~/.claude/claude_desktop_config.json that registers the Probe MCP server (see the README for the exact snippet).

---

Supported languages (as listed)

Rust, JavaScript/JSX, TypeScript/TSX, Python, Go, C/C++, Java, Ruby, PHP, Swift, Solidity, Crystal, C#, Bash, QML, Markdown, HTML, YAML, and others via Tree‑sitter.

---

Installation options

Method	Command
npm (global)	npm install -g @probelabs/probe
curl (macOS/Linux)	`curl -fsSL https://raw.githubusercontent.com/probelabs/probe/main/install.sh
PowerShell (Windows)	`iwr -useb https://raw.githubusercontent.com/probelabs/probe/main/install.ps1
From source	git clone https://github.com/probelabs/probe.git && cd probe && cargo build --release && cargo install --path .

---

License

The repository states a License section but does not specify the exact SPDX identifier in the README. Check the LICENSE file in the repo for the precise terms.

---

Where to learn more

• Full docs: https://probelabs.com/probe
• CLI reference: docs/probe-cli/*
• Agent SDK: docs/probe-agent/sdk/*
• Community: GitHub Issues and Discord (links in the README).

---

All information above is taken directly from the project's README; no external assumptions have been added.