Repo to LLM Context Packer
Turn a codebase into one clean file for ChatGPT, Claude or Gemini. Drop a folder or ZIP, or paste a public GitHub URL: it skips junk and lockfiles, redacts secrets, shows token counts per file against your context window, and exports Markdown, XML or plain text, optionally split into parts.
GitHub URLs use GitHub’s public API (60 anonymous requests per hour; one pack uses two) and download files one by one. For private repositories or very large ones, download the ZIP from GitHub and drop it here.
Filters & options
| File (largest first) | Tokens | Size |
|---|
Paste the output into ChatGPT, Claude or Gemini, or attach the downloaded file. Always check the result before sharing code you do not own.
How to give an AI your whole codebase
Chat assistants cannot browse your repository, so the usual way to ask questions about a project is to paste its code. Doing that by hand wastes the context window on lockfiles, build output and images and risks leaking secrets. The packer builds one file with a directory tree and each source file in a clearly labelled block, so the model can see how the project fits together.
What gets left out, and why
- Junk:
node_modules,.git,dist,build, caches, lockfiles, minified files, source maps, images, fonts, archives and other binaries add tokens without helping. - Secrets:
.envfiles, private keys and credential files are skipped, and API keys, tokens, private-key blocks and passwords found in code are replaced with[REDACTED:type]. - Everything your .gitignore ignores, plus any patterns you add. Use “Only include” to pack just one folder or language.
Fitting the context window
Token counts are shown per file and in total against the budget you choose (an estimate by default, exact for OpenAI models). If the pack is too large, exclude the biggest files, narrow it to the folders that matter, or split it into parts that each fit.
Markdown, XML or plain text?
Markdown with fenced code blocks is the most readable and works everywhere. XML tags give models an unambiguous boundary around each file, which some prompts prefer. Plain text is the most compact.
How to use the Repo Packer
- Drop a folder or ZIP, or paste a public GitHub URL.
- Check the token total against your context budget and untick files you do not need.
- Choose Markdown, XML or plain text, and split into parts if needed.
- Copy the result or download it and give it to your AI.
Why use this packer
- Skips junk, lockfiles and binaries automatically
- Redacts API keys, tokens and passwords
- Token counts per file, exact for GPT models
- Works from a folder, ZIP or GitHub URL, all in your browser
Related tools
Frequently Asked Questions
How do I give ChatGPT or Claude my whole codebase?
Chat assistants cannot read your repository directly, so pack the code into one file: drop your project folder or a ZIP onto this page, or paste a public GitHub URL. Check the token total against the model's context window, then copy the output into the chat or attach the downloaded file.
Is my code uploaded anywhere?
No. Folders and ZIP files are read and packed in your browser and never sent to a server. When you paste a GitHub URL, your browser downloads the public files directly from GitHub; nothing is sent to Brace.
Does it remove secrets?
It skips files that normally hold secrets (.env files, private keys, credential files) and replaces API keys, tokens, private-key blocks and passwords found in code with [REDACTED:type]. Detection is pattern-based and cannot catch every secret, so review the output before sharing proprietary code.
How many tokens will my repository use?
The tool shows a total and a per-file count. The default is a fast estimate of about four characters per token; choose a GPT tokenizer for exact counts. Claude and Gemini counts are estimates. If the total is over your budget, exclude the largest files, narrow the include list, or split the output into parts.
Can it pack private repositories or very large repos?
From a URL it works with public repositories only and is limited by GitHub's anonymous rate limit of 60 requests per hour. For private or very large repositories, use GitHub's Download ZIP (or zip the folder yourself) and drop it onto the page.
Which output format is best: Markdown, XML or plain text?
Markdown with fenced code blocks is the most readable and works with every assistant. XML tags put an unambiguous boundary around each file, which some prompts prefer. Plain text is the most compact. All three include a directory tree.