TokenTrim

Trim tokens. Cut costs. Code smarter.

Stop overpaying for LLM APIs. TokenTrim optimizes your prompts locally right in VS Code with dual optimization modes: instant rule-based compression or advanced ML-powered optimization.

✨ Features

Core Features

• 📊 Token Counting: Real-time token count using tiktoken (OpenAI), approximations for Claude/Gemini
• 💰 Cost Estimation: Accurate cost calculation across 9+ LLM models
• ✂️ Dual Optimization Modes:
  • Quick Compress (Rule-Based): 30-50% reduction, instant
  • ML Optimize (LLMLingua): 50-80% reduction, 2-3 seconds
• 🌐 Multi-Provider Support: OpenAI, Anthropic Claude, Google Gemini
• 🔒 Privacy-First: 100% local processing, no data sent to external servers

Advanced Features (v2.0)

• 🤖 LLMLingua Integration: State-of-the-art ML compression
• 📊 Multi-Model Comparison: Compare costs across all supported models
• 🔄 Intelligent Fallbacks: Automatic fallback to rule-based if ML unavailable
• ⏱️ Progress Indicators: Real-time feedback during ML compression

🚀 Quick Start

1. Select text in any editor
2. Open the TokenTrim sidebar (or check the status bar for token count)
3. Choose your optimization mode:
   • Click "Quick Compress" for instant results
   • Click "ML Optimize" for maximum compression
4. Review the before/after comparison
5. Click "Replace Selection" to save tokens and money

⌨️ Keyboard Shortcuts

• Ctrl+Shift+O (Windows/Linux) or Cmd+Shift+O (Mac): Optimize with rules
• Ctrl+Shift+M (Windows/Linux) or Cmd+Shift+M (Mac): Optimize with ML

📋 Commands

• TokenTrim: Optimize Selected Text - Rule-based optimization
• TokenTrim: Optimize with ML (LLMLingua) - ML-powered compression
• TokenTrim: Show Sidebar
• TokenTrim: Select Model

Supported Models

18 models across 3 providers:

OpenAI (5 models):

• GPT-4 Turbo
• GPT-4
• GPT-4 32K
• GPT-3.5 Turbo
• GPT-3.5 Turbo 16K

Anthropic Claude (7 models):

• Claude Opus 4.5
• Claude Sonnet 4.5
• Claude 3.5 Sonnet
• Claude 3.5 Haiku
• Claude 3 Opus
• Claude 3 Sonnet
• Claude 3 Haiku

Google Gemini (6 models):

• Gemini 3 Pro
• Gemini 3 Flash
• Gemini 2.0 Flash
• Gemini 1.5 Pro
• Gemini 1.5 Flash
• Gemini 1.5 Flash-8B

Note: Claude and Gemini token counts are approximations (±5% accuracy).

🤖 LLMLingua Setup (Optional)

For advanced ML compression, install LLMLingua:

Prerequisites

• Python 3.8 or higher
• pip (Python package manager)

Installation

1. The extension will prompt you to install LLMLingua on first use
2. Click "Install" to automatically install dependencies
3. Wait 2-3 minutes for model download (~500MB)

Manual Installation

pip install llmlingua torch transformers

Troubleshooting

• Python not found: Install from python.org
• Installation fails: Try pip install --upgrade pip first
• Model download slow: Use a VPN if in restricted regions

Note: LLMLingua is optional. Rule-based optimization works without it.

🛠️ How It Works

Rule-Based Optimization (Quick Compress)

• ✂️ Whitespace normalization - Remove excessive spacing
• 📝 Markdown stripping - Clean unnecessary formatting
• 🎯 Phrase condensing - Replace verbose phrases with concise alternatives
• 🧹 Filler removal - Strip words that don't add value
• ⚡ Punctuation optimization - Simplify without losing meaning

ML Optimization (LLMLingua)

• 🧠 Semantic compression - Preserves meaning while removing redundancy
• 📉 Token-level optimization - Intelligent token selection
• 🎯 Context-aware - Understands prompt structure
• ⚡ 50-80% reduction - Superior compression ratios

All transformations preserve semantic meaning.

📦 Requirements

• VS Code 1.85.0 or higher
• Python 3.8+ (optional, for ML optimization)

📄 License

MIT - Free forever, open source

---

Made with ❤️ for developers who care about LLM costs