Probe Context retrieval that understands code
Most tools do text search + a few surrounding lines. Probe understands functions, symbols, and document structure, ranks results, and extracts only the relevant scope.
$ probe "auth flow" ./srcAST parsing → Symbol matchingBM25 ranking → Scope detectionWhy Most "AI Code Understanding" Breaks at Scale
In small repos, "search a few strings and read 20 lines around it" works. In large codebases it fails because understanding requires structure, boundaries, and the big picture—not just matching text.
- Code isn't text. It's symbols, scopes, call paths, and ownership boundaries.
- Docs aren't text. They're sections, headings, and contracts.
- The failure mode is predictable: tools say "I understand" too early.
Probe is built to make "I understand" mean something.
How Probe Works
Structure-aware retrieval
Search like you understand the language, not like you're scanning strings
- Understands functions, variables, and scopes (not just matching lines)
- Markdown-aware: respects headers, sections, and document structure
- Returns results with file paths + line numbers + clear boundaries
Elastic depth
Start narrow, expand only when needed
- Extracts only the scope you need (signature / key lines / relevant block)
- Shows where a function starts and ends, so the agent can decide to expand
- "Go deeper" is deliberate: pull the full function body only when required
Stop shipping entire files into prompts just to answer one question.
Precise extraction
Precision tools beat "read the whole file"
- Extract by line range, symbol/function name, or section
- Produce consistent "context artifacts" that workflows can reuse
- Token-efficient: one Probe query can replace many "search/open/scroll" steps
Better retrieval → better plans → better automation.
Get Started in Seconds
Choose your preferred way to integrate Probe into your workflow.
Command Line
Search any codebase instantly with a single command. No installation required - just run via npx.
AI Chat Mode
Interactive chat interface for exploring code with AI assistance. Opens in your browser.
MCP Server
Connect to Claude Code, Cursor, Codex, or any MCP-compatible AI editor.
Node.js SDK
Programmatic access for building custom integrations and workflows.
Code as Text vs. Code as a Codebase
Common approach
- Run 2–3 text queries
- Grab a few lines around matches
- Assume it's enough context
- Move on confidently (often wrong)
Probe approach
- Build a ranked set of structural hits
- Extract scoped, minimal context with boundaries
- Highlight the important identifiers
- Expand intentionally when deeper context is needed
It's not about "more context." It's about the right context, delivered efficiently.
Powerful Query Syntax—Without Heavyweight Indexing
Probe supports advanced, search-engine style queries and runs fast on large repos without requiring an external indexing service.
Familiar Query Operators
Boolean operators, phrase matching, and fuzzy search—a query syntax you already know.
Code + Markdown
Works across source files and documentation with structural awareness for both.
Local / On-Prem Ready
Designed for use in real enterprise repos. Retrieval runs locally; you control what context is sent to the model.
50+ Languages
Support for all major programming languages with accurate AST parsing.
MCP + SDK
Built-in MCP server for Claude and Cursor. Node.js SDK for custom integrations.
Open Source
Apache 2.0 licensed. Run it, modify it, embed it in your workflows.
Probe makes context retrieval deterministic: scoped, ranked, expandable.
Turn Your Repo Into a Queryable Knowledge System
Open source and free to use. Single binary, no dependencies.