repopack

Pack an entire codebase into a single LLM-optimized markdown file.

Install

npm install -g repopack

Or run without installing:

npx repopack

Requires Node.js >= 18.

Usage

# Pack current directory → repopack-output.md
npx repopack

# Pack a specific directory with a custom output file
npx repopack ./src -o context.md

# Truncate largest files first to stay under a token budget
npx repopack --max-tokens 100000

# Only include TypeScript files, skip tests
npx repopack --include "*.ts,*.tsx" --exclude "*.test.*"

# Strip comments and collapse blank lines (shrinks token count)
npx repopack --compress

# Print token stats without writing any file
npx repopack --stats

# Copy output to clipboard instead of writing a file
npx repopack --copy

All options

| Flag | Default | Description | |------|---------|-------------| | [dir] | . | Directory to pack | | -o, --output <file> | repopack-output.md | Output file path | | --include <globs> | **/* | Comma-separated glob patterns to include | | --exclude <globs> | — | Comma-separated glob patterns to exclude | | --max-tokens <n> | — | Token budget; drops largest files first until under limit | | --compress | off | Strip comments and collapse blank lines | | --stats | off | Print token stats to stdout, skip file write | | --copy | off | Copy output to clipboard instead of writing a file |

Output format

# Repository: my-project
Generated by repopack | 42 files | ~18,400 tokens

## File Tree

my-project ├── src │ ├── index.ts │ └── utils.ts └── package.json

## src/index.ts
```typescript
// ... file contents

src/utils.ts

// ... file contents

Each file gets a level-2 heading with its relative path and a fenced code block with the correct language tag. Paste the whole file directly into any chat or API call.

---

## Smart defaults

The following are excluded automatically, with no configuration needed:

- `node_modules/`
- `.git/`
- Lock files: `package-lock.json`, `yarn.lock`, `pnpm-lock.yaml`, `bun.lockb`
- Environment files: `.env`, `.env.*`
- Binary files: images, fonts, audio, video, archives, executables, PDFs

Your `.gitignore` rules are also respected automatically.

---

## Token counting

repopack uses `cl100k_base` (tiktoken) to count tokens — the same encoding used by GPT-4. Claude and Gemini token counts are approximated with the same encoder, which is accurate to within ~5% for typical source code.

`--stats` output shows usage against each model's context window:

Total tokens: 18,400

Model context usage: GPT-4 18,400 / 128,000 tokens (14.4%) Claude 18,400 / 200,000 tokens (9.2%) Gemini 18,400 / 1,000,000 tokens (1.8%)

**Context limits used for comparison:**

| Model | Limit |
|-------|-------|
| GPT-4 | 128,000 tokens |
| Claude | 200,000 tokens |
| Gemini | 1,000,000 tokens |

Use `--max-tokens` to truncate when you need to stay within a specific limit. Files are dropped largest-first until the output fits.

---

## .repopackignore

Place a `.repopackignore` file in your project root to exclude files that you don't want in `.gitignore`. Uses the same syntax as `.gitignore`.

.repopackignore

dist/ *.generated.ts fixtures/

Priority order (highest to lowest): `--exclude` flag → `.repopackignore` → `.gitignore` → built-in defaults.

---

## Compression

`--compress` reduces token count by:

1. Stripping comments (JS/TS/Python; other languages are left untouched)
2. Collapsing consecutive blank lines to a single blank line
3. Trimming trailing whitespace per line

Typical reduction is 10–30% depending on how comment-heavy the codebase is.

---

## License

MIT — see [LICENSE](LICENSE).