zvec-ai/zvec-grep

What it solves

zvec-grep (zg) provides a unified, local-first search layer that allows both humans and AI agents to find information across source code, documents, and structured data. It solves the problem of inefficient retrieval in large workspaces where keyword searches (like grep) are too rigid and pure semantic searches can be imprecise, reducing the number of tool calls and tokens an agent needs to find the right evidence.

How it works

It unifies three different search methodologies into a single interface:

1. Ripgrep: For exact text and regex matching.
2. BM25: For traditional lexical ranking.
3. Vector Search: For semantic discovery based on meaning.

Users can index a workspace locally, and the tool can be used via a CLI or integrated into AI agents (such as Claude Code or OpenCode) via the Model Context Protocol (MCP). Data and indexes remain on the local machine by default.

Who it’s for

• Developers who want a powerful, multi-modal search tool for their local files.
• AI Agent creators who want to provide their agents with a high-precision retrieval tool to navigate complex codebases or document sets.
• Users prioritizing data privacy who require local-first indexing and embedding models.

Highlights

• Hybrid Search: Combines semantic, lexical, and regex search to balance discovery and precision.
• Agent-Ready: Built-in integrations for agents to autonomously decide when and how to search.
• Local-First: Keeps files, indexes, and models on the user's machine.
• Multi-Format Support: Handles source code and prose while preserving structure and source locations.
• Performance Optimized: Reduces model context noise and tool-call frequency for agents.