Qwen Developers Open-Source zg, a Unified Local Search Tool for Agents

Coding agents often expend a significant portion of their tool budget on search operations. [1] While ripgrep handles known symbols exactly, keyword matching can fail for plain language descriptions, causing agents to waste tool calls and tokens guessing terms or reading entire files. [1] To address this, the Qwen Developer team announced zg (zvec-grep), an open-source local-first search layer. [1] This tool places semantic search, BM25, and ripgrep behind one interface intended for both humans and agents. [1]

The zg tool is deployable immediately, installing from npm as `@zvec/zvec-grep`. [1] It requires Node.js 22 or newer on macOS, Linux, or Windows, operates without a GPU using the default model, and is released under the Apache 2.0 license. [1] zg indexes a workspace into an anonymous directory located at `<root>/.zvec-grep/` and offers four retrieval routes. [1] These include a hybrid default, `--fts` for BM25-ranked exact terms, `--vector` for conceptual similarity, and `--rg` for exhaustive literal or regex matching. [1]

By consolidating disparate search methodologies—exact match, lexical, and semantic—into a single local tool, zg directly targets a major bottleneck for coding agents: context assembly. Unlike relying solely on cloud-based vector databases or standalone ripgrep commands, this unified approach allows agents to toggle between broad conceptual searches and precise regex matches without context switching. The choice of an Apache 2.0 license and a no-GPU default model positions it for frictionless integration into existing developer workflows and agentic pipelines. What the sources don't address: how zg's indexing performance and latency scale when applied to extremely large, enterprise-scale codebases.