If you’ve ever needed to give an LLM context for multiple files as you code, then you know how tedious it is to copy and paste all of the files into the LLM window.

This Visual Studio Code extension lets you compile all of your files into a single .txt file.

Using glob patterns you can include or exclude files that wouldn’t be helpful for an LLM context (such as build files and libraries).

GitHub Repo

📄 flatten-repo

Publish Extension

Flatten your entire codebase into clean, readable .txt files — optimized for LLMs like ChatGPT, Claude, and Gemini.

✨ Features (v0.0.12)

  • 🔁 Auto-flattens your workspace into plain .txt chunks
  • 🧠 Built for LLM parsing, prompt engineering, and static code analysis
  • 📂 Each chunk includes a directory tree overview of included files
  • ✂️ Auto-chunks content using a configurable token limit (~4 characters per token)
  • 🔍 Powerful support for glob-based ignore, whitelist, and blacklist
  • 🧾 All configs live in a single .flatten_ignore file (generated automatically)
  • ⚙️ Customize file extensions, folders to ignore, and token size limits
  • 📁 Output saved in timestamped files under /flattened
  • 🚫 Auto-adds /flattened to your .gitignore

Each chunk starts with a tree-like outline of included files, followed by:

=== FILE: path/to/file.ext ===

⚙️ How to Use

  1. Open a folder in VS Code
  2. Open Command Palette (Cmd+Shift+P / Ctrl+Shift+P)
  3. Run: Flatten Project to TXT
  4. View flattened files inside the /flattened folder

🛠️ Configuration

You can configure behavior in your .vscode/settings.json:

"flattenRepo.includeExtensions": [".ts", ".tsx", ".js", ".jsx", ".py", ".html", ".css"],
"flattenRepo.ignoreDirs": ["node_modules", ".git", "dist"],
"flattenRepo.maxChunkSize": 200000

Or configure per-project settings via .flatten_ignore.

📄 .flatten_ignore

This single file controls:

  • ✅ Glob-based global ignore rules
  • ➕ Optional whitelist or blacklist
  • 📐 Token limits via a settings: section

Auto-generated in /flattened if missing.

🔁 Sample .flatten_ignore

# Ignore rules
global:
node_modules
.git
dist

# Whitelist (optional)
whitelist:
src/**/*.js

# Blacklist (optional)
blacklist:
**/*.test.js
.env

# Settings (optional)
settings:
maxTokenLimit: 50000
maxTokensPerFile: 25000

# Suggestions:
#   Claude 3.7: 128k tokens
#   ChatGPT 4o: 128k tokens
#   ChatGPT o3-mini-high: 200k tokens
#   Claude 2: 100k tokens
#   Anthropic Claude 3 Opus: 200k tokens
#   Cohere Command: 32k tokens
#   Google PaLM 2: 8k tokens
#   Meta LLaMA 2: 4k tokens

📐 Output Format

Each .txt output file looks like this:

=== Directory Tree ===
├─ App.tsx
├─ index.js
└─ components
   ├─ Header.tsx
   └─ Footer.tsx

=== FILE: App.tsx ===
import React from 'react';
...

✅ Use Cases

  • Preparing source code for LLM input
  • Clean context formatting for ChatGPT, Claude, Gemini, etc.
  • Snapshotting your repo for AI audits or static reviews
  • Prompt engineering pipelines
  • Code flattening for full-project memory with agents

🐞 Known Limitations

  • No graphical UI (yet) — command-only
  • Does not flatten binary or image files
  • Some advanced glob edge cases may need refinement

🧪 Contributing

Want to help improve this tool?

Ideas to explore:

  • File token counts
  • Markdown formatting
  • Multi-model export formats
  • UI interface for selecting flatten options

🔗 Resources

Made with ❤️ to help devs and LLMs speak the same language.