zg(zvec-grep) Open-Sourced: Hybrid Local Search Infrastructure for AI Agents

When AI agents perform complex tasks, natural language-based intent recognition and accurate information retrieval have been identified as limitations of traditional keyword search. To address these issues, zg(zvec-grep) has been open-sourced. zg is a local-first search infrastructure that combines Zvec's vector search and BM25 capabilities with ripgrep(rg) to extract and organize information from code and documents.

Key Features and How It Works

• Hybrid Search: Combines semantic search, BM25, and rg to simultaneously provide the precision of keyword matching and the flexibility of natural language understanding.
• Context Efficiency: Fuses and ranks results from multiple search paths, helping agents reduce unnecessary token consumption and repetitive searches while quickly accessing relevant information.
• Support for Various Formats: Supports diverse file formats including code, documents, and structured data, preserving symbols, headings, and hierarchical structures to expand the searchable scope.
• Local-First Design: File scanning, indexing, and embedding generation run locally by default, protecting data privacy.

Adoption and Integration

zg supports macOS, Linux, and Windows, providing CLI and MCP(Model Context Protocol) interfaces. It can be installed using the command npm install -g @zvec/zvec-grep, and automatically detects major AI coding tools such as Codex, Claude Code, Cursor, and OpenCode to complete MCP configuration. This allows developers and AI agents to immediately utilize hybrid search capabilities without complex integration setups.