EnviroLLM CLI

A command-line tool for benchmarking energy consumption and performance of local LLMs across Ollama, LM Studio, vLLM, and other platforms.

Installation

Run with npx (no installation needed):

npx envirollm start

Or install globally:

npm install -g envirollm

Requirements

• Node.js 14+
• Python 3.7+
• pip (Python package manager)

Quick Start

1. Start the backend service:

   npx envirollm start

2. Run benchmarks:

   # Ollama models
   npx envirollm benchmark --models llama3:8b,phi3:mini
   
   # LM Studio or other APIs
   npx envirollm benchmark-openai --url http://localhost:1234/v1 --model llama-3-8b

3. View results:

   • CLI: Results displayed in terminal after benchmark completes
   • Web: envirollm.com/optimize for more detailed UI

Commands

# Service Management
npx envirollm start              # Start backend service (required for benchmarks)
npx envirollm start --port 8002  # Start on custom port
npx envirollm status             # Check if service is running

# Benchmarking - Ollama
npx envirollm benchmark --models llama3:8b,phi3:mini
npx envirollm benchmark --models llama3:8b --prompt "Write a sorting function"
npx envirollm benchmark --models llama3:8b,llama3:8b-q8  # Compare quantizations

# Benchmarking - LM Studio, vLLM, Custom APIs
npx envirollm benchmark-openai --url http://localhost:1234/v1 --model llama-3-8b
npx envirollm benchmark-openai --url http://localhost:8000/v1 --model meta-llama/Llama-2-7b-hf
npx envirollm benchmark-openai --url http://localhost:1234/v1 --model phi-3 --prompt "Custom prompt"
npx envirollm benchmark-openai --url http://localhost:1234/v1 --model llama-3-8b --api-key your-key

# Data Management
npx envirollm clean              # Remove all stored benchmark data

# Process Monitoring
npx envirollm detect             # List detected LLM processes
npx envirollm track --auto       # Auto-detect and track LLM processes
npx envirollm track -p python    # Track specific process by name

Benchmarking Details

Requirements:

• Ollama: Install Ollama and run ollama serve
• LM Studio/vLLM/Custom: API must be running on specified URL

Metrics Collected:

• Energy consumption (Wh)
• Tokens per second
• CPU/GPU/memory usage
• Quantization detection (Q4, Q8, FP16)
• Power draw (W)
• Response quality comparison

Data Storage: All benchmark results are stored locally at ~/.envirollm/benchmarks.db. Your data never leaves your machine.

Web Interface Alternative

You can also run benchmarks using the web interface at envirollm.com/optimize after starting the monitoring service with npx envirollm start. The web UI provides:

• Visual model selection for Ollama, LM Studio, and custom APIs
• CSV export functionality for benchmark data
• Response comparison view to evaluate output quality
• Custom prompt configuration
• Same backend - results sync between CLI and web

Features

• Real-time Monitoring: CPU, GPU, memory, and power consumption
• Multi-Platform Benchmarking: Support for Ollama, LM Studio, vLLM, and OpenAI-compatible APIs
• Optimization Recommendations: System-specific suggestions for reducing energy usage
• Process Detection: Automatic identification of running LLM processes

How It Works

The CLI starts a local Python backend service that collects system metrics using psutil and pynvml. The web dashboard at envirollm.com/dashboard automatically detects if you're running the local service and switches to display your real hardware metrics instead of demo data.

Benchmarking through the CLI or at envirollm.com/optimize runs inference requests against your local models while monitoring energy consumption, token generation speed, and resource usage in real-time.

License

MIT