Qwen developers launch local-first search tool unifying multiple search methods

The Qwen Developer team has released zg (zvec-grep), an open-source search tool integrating semantic search, BM25, and ripgrep under Apache 2.0 on GitHub. Designed for human users and coding agents, zg offers a local-first approach with hybrid, BM25, vector-based, and ripgrep modes. It indexes workspaces, provides freshness indicators, and supports local and remote embedding models, with remote access requiring explicit authorization. zg improves efficiency and accuracy in benchmark tests by reducing tool calls and input tokens. Deployable via npm with Node.js 22 or newer, it excludes .git and .zvec-grep directories, allows incremental reindexing, and limits default agent toolset to two functions. A full toolset is available via --mcp-toolset, but agents must not silently modify persistent indexes. The embedding catalog includes ten local models and three remote Qwen endpoints, with Model2Vec as the default. Remote use requires explicit configuration or grants.