Mgrep

✨Key Features

• ▸Natural Language Queries: Search files semantically in any language (e.g., "where do we set up auth?") across code, text, PDFs, or images; returns ranked results with context like line ranges or page numbers.
• ▸Background Indexing: Use mgrep watch to automatically index git repos (ignoring .gitignore/.mgrepignore), upload to a cloud store, and keep it synced for fast, always-up-to-date searches.
• ▸Web Search Integration: Add --web flag to blend local results with online sources, with optional reranking for relevance; ideal for research or troubleshooting (e.g., "best practices for TypeScript error handling").
• ▸Output Options: Customize with flags like -m for result limits, -c to show content snippets, or -a for AI-summarized answers; supports dry runs and max file size limits (default 10MB).
• ▸Agent Compatibility: Integrates with tools like Claude Code, OpenCode, and Codex via install commands; reduces token usage by ~2x in benchmarks compared to traditional grep workflows.
• ▸Multimodal Support: Handles text/code/PDFs/images now; audio/video planned; multilingual embeddings via Mixedbread Search SDK.
• ▸Easy Setup and Config: Install via npm/pnpm/bun; login with API key; config via CLI flags, env vars, or YAML files (.mgreprc.yaml); Apache-2.0 licensed.
• ▸Performance Benchmarks: In a 50-task QA test, mgrep + Claude Code achieved similar/better quality with half the tokens; demo available online.
• ▸Development and Community: TypeScript-based with 2.2k GitHub stars; 8 contributors; active updates (latest v0.1.8 on Dec 15, 2025); join Mixedbread Slack for support.