Combine files for AI merges a project's source files into one bundle you can paste into ChatGPT, Claude or Gemini. Drop files or a whole folder; dependency and build folders, binaries and very large files are skipped automatically. Each file is labelled with its path inside a code fence or an XML tag, a file tree goes at the top, and a running token estimate shows whether it fits the model's context window. Files are read in your browser and never uploaded.
| Tokens | About |
|---|---|
| 8,000 | 30 KB of code |
| 32,000 | 125 KB |
| 128,000 | 500 KB |
| 200,000 | 800 KB |
| 1,000,000 | 4 MB |
Anything in node_modules, .git, dist, build and similar folders, binary files such as images, and files over 2 MB.
Both work everywhere. XML <file> tags are what Anthropic recommends for Claude; Markdown fences read more naturally in ChatGPT.
It's an estimate at about four characters per token. Paste the bundle into the LLM Token Counter for an exact count per model.
Yes. Click an extension to exclude it, or remove single files and reorder the rest.
No. They're read in your browser. Check for secrets such as .env files before you share the bundle with an AI service.
Files and folders are read in the browser. Paths under node_modules, .git, dist, build, .next, __pycache__, virtual environments, target and coverage are skipped, as are files over 2 MB and binary files — detected by a NUL byte or a high share of control characters in the first 8 KB, the same heuristic git uses. Each remaining file becomes a section: in Markdown mode a heading with its path and a code fence tagged with its language (from the extension, or names such as Dockerfile and Makefile), with the fence made one backtick longer than any run of backticks inside the file so Markdown files containing code blocks don't break the bundle; in XML mode a <file path="…"> element with the path attribute-escaped, which Anthropic recommends for long documents. An optional file tree drawn with box characters goes at the top, blank-line runs can be collapsed and trailing spaces trimmed to save tokens. The size is estimated at four characters per token and compared against the selected model's context window.