🏹 Robin Wood

Stealing from the rich (LLM providers) to give to the poor (you)

Your LLM bill is highway robbery. Audio files with 2 minutes of silence? $$$ Text with unnecessary whitespace? $$$ Images at 4K when the model can't even see that well? $$$

Robin Wood compresses your audio, text, and images before you send them to OpenAI, Anthropic, or whoever is charging you $20/million tokens.

We steal bytes. You save money. Sam Altman cries.

💸 Why you need this

• That 5-minute voice memo? 70% silence. You're paying for air.
• Your markdown? Full of \n\n\n\n. You're paying for nothing.
• That screenshot? 4K. The AI doesn't care.

LLM providers charge per token/byte. Robin Wood cuts the fat. Keep the signal, drop the noise.

🚀 Installation

npm install robin-wood

🎯 Usage (it's dead simple)

import steal from 'robin-wood';

// Audio: Remove silence, speed up, normalize
const audio = await steal
  .audio(yourAudioBuffer)
  .removeSilence(-40, 100)  // Remove silence below -40dB
  .speedup(1.2)              // 20% faster
  .normalize()               // Balance volume
  .run();

console.log(`Saved ${audio.duration}s of your money`);

// Text: Trim whitespace, minify
const text = await steal
  .text("  Too    many   spaces    ")
  .trim()
  .minify()
  .run();

console.log(`Compression ratio: ${text.ratio}x`);

// Image: Resize, compress
const image = await steal
  .image(bigAssImage)
  .resize(800, 600)
  .compress(85)
  .run();

🎪 What it does

Audio Pipeline

• speedup(rate): Make it faster without chipmunk effect (1.1x - 2x recommended)
• removeSilence(): Delete awkward pauses. Save tokens. Save face.
• normalize(): Balance audio levels
• volume(level): Adjust volume (0.0 - 2.0)

Text Pipeline

• trim(): Kill unnecessary whitespace
• minify(): Compress text structure
• compress(): gzip/brotli compression

Image Pipeline

• resize(w, h): LLMs don't need your 4K screenshot
• compress(quality): JPEG compression (0-100)

💡 Real Talk

This library doesn't do magic. It does the obvious optimizations you should've been doing anyway:

1. ✂️ Removes silence from audio (why pay for dead air?)
2. 🗜️ Compresses text (LLMs don't care about your double newlines)
3. 📉 Downsizes images (GPT-4 Vision has a resolution limit anyway)

Result? 30-70% cost reduction on typical payloads. YMMV.

🏹 Philosophy

We named it Robin Wood because:

1. Robin Hood stole from the rich to give to the poor
2. We steal bytes from bloated files to save your money
3. Wood sounds like "would" ("Robin would save you money")
4. We thought it was funny at 2 AM

The main function is called steal(). Because that's what we do.

📚 API Reference

Audio

steal.audio(buffer: Buffer | ArrayBuffer | Uint8Array)
  .speedup(rate: number)              // 0.1 - 4.0
  .removeSilence(thresholdDb, minMs)  // Default: -40dB, 100ms
  .normalize()                        // Balance volume
  .volume(level: number)              // 0.0 - 2.0
  .run()

Important: Pass raw PCM Float32 data, not MP3/WAV files. Decode first.

Returns: { data: Buffer, duration: number, sampleRate: number }

Text

steal.text(string)
  .trim()                    // Remove whitespace
  .minify()                  // Minify structure  
  .compress('gzip' | 'brotli')
  .run()

Returns: { data: string, ratio: number }

Image

steal.image(buffer)
  .resize(width, height)
  .compress(quality: number)  // 0-100
  .run()

Returns: { data: Buffer, width: number, height: number }

🤔 FAQ

Q: Will this break my audio/text/images?
A: No. Everything is lossless except where you explicitly compress. Test before production.

Q: How much will I save?
A: Depends on your data. Typical savings: 30-70% on audio with silence, 10-30% on text, 50-80% on images.

Q: Is this just a wrapper around ffmpeg?
A: No. Pure JavaScript/TypeScript. No external dependencies. Runs anywhere Node runs.

Q: Can I use this with OpenAI/Anthropic/etc?
A: Yes. Process your data with Robin Wood, then send to any LLM API.

Q: Does this work in the browser?
A: Not yet. Node.js only for now. Browser support coming soon™.

🎯 Who this is for

• Developers sending audio to Whisper API
• Apps using GPT-4 Vision with user-uploaded images
• Anyone paying $20/million tokens and crying about it
• People who like stealing from billionaires (legally)

🚫 Who this is NOT for

• People who don't care about LLM costs (lucky you)
• Apps that need lossless audio (use FLAC)
• Anyone who thinks "optimize" is a buzzword

📦 Requirements

• Node.js >= 16
• A burning desire to pay less money

🤝 Contributing

PRs welcome. Add more compression strategies. Make it faster. Make Sam Altman's yacht smaller.

📄 License

MIT - Steal freely (see what we did there?)

🔗 Links

• NPM
• GitHub
• Issues (tell us what broke)

Made by developers tired of paying $200/month for API calls 🏹

"In Sherwood Forest, we compressed the Sheriff's taxes. In Silicon Valley, we compress your tokens."