// the find
yamadashy/repomix
📦 Repomix is a powerful tool that packs your entire repository into a single, AI-friendly file. Perfect for when you need to feed your codebase to Large Language Models (LLMs) or other AI tools like Claude, ChatGPT, DeepSeek, Perplexity, Gemini, Gemma, Llama, Grok, and more.
Repomix concatenates an entire codebase into a single file (XML, Markdown, JSON, or plain text) for feeding to LLMs. It handles gitignore rules, token counting, secret scanning via Secretlint, and can clone remote repos on the fly. Aimed at developers who want to do code review, documentation generation, or Q&A against a full codebase without manually copying files.
1. The --compress flag uses Tree-sitter to extract function/class signatures while stripping bodies — a genuine 70% token reduction that preserves enough structure for most analysis tasks, not just naive truncation. 2. Built-in Secretlint integration means you won't accidentally dump an API key into your LLM context; it catches patterns before the output file is written. 3. MCP server mode (--mcp) lets AI assistants call pack_codebase or grep_repomix_output directly, so the 'copy this file into the chat' step disappears from the workflow. 4. --token-budget with a non-zero exit code on overflow is a proper CI guard — stops agent pipelines from silently sending 500k-token blobs to models with 200k context limits.
1. The default XML format breaks if your code contains XML-like strings or CDATA — they added --parsable-style to escape it, but that's a band-aid on a structural choice that was wrong from the start; Markdown or plain text formats don't have this problem. 2. --compress is still experimental and Tree-sitter grammar coverage is uneven across languages; it works well for TypeScript/JavaScript but you can get incomplete or malformed output for less common languages without any warning. 3. --remote only works for public repos — private repos require a local clone first, which the README mentions only in passing, making it a surprise when you try it on a private org repo. 4. For genuinely large monorepos, even compressed output regularly hits 200k+ tokens; the tool gives you --token-count-tree to see the damage but no automated suggestions for which directories to cut — that judgment call is still entirely on you.