greenprompt 0.0.1

A green prompt tool to estimate energy usage of LLM prompts

GreenPrompt 🌱

Overview

GreenPrompt is a local-first tool that tracks and visualizes the real-world energy consumption of AI prompts, locally. It intercepts or wraps LLM calls (via Ollama’s local API), samples CPU/GPU power, logs prompt metadata and energy usage, and provides both CLI and Web UI interfaces for analysis.

macOS only (uses powermetrics for power sampling)

Features

• Baseline & Prompt Sampling: Continuously samples system power in a 10‑minute sliding window; computes a 1‑minute baseline before each prompt and measures power during execution.
• Token & Model Tracking: Captures prompt and completion token counts, model used.
• Energy Estimation: Calculates energy in Wh based on sampled power data and execution duration.
• Persistent Logging: Stores every prompt run in a SQLite database (greenprompt_usage.db).
• CLI Interface: Run prompts, view immediate stats.
• Web API & Dashboard: Flask-based endpoints for retrieving usage and a simple HTML dashboard.
• Proxy Server: (Optional) Intercept existing Ollama API traffic to log usage without code changes.

Architecture & Flow

1. Power Sampling
   • samplerMac.py defines PowerMonitor, a background thread that samples combined CPU/GPU power every second and retains a 10‑minute history.
2. Prompt Execution
   • core.py’s run_prompt function:
     • Retrieves a 1‑minute baseline average from PowerMonitor.
     • Sends the user prompt to Ollama at 127.0.0.1:5000:11434 (or via proxy).
     • Records start and end timestamps.
     • Queries PowerMonitor for samples during execution.
     • Calculates average power and energy.
3. Data Persistence
   • dbconn.py provides init_db() and save_prompt_usage(), storing each run’s metadata in SQLite.
4. Interfaces
   • CLI (cli.py): greenprompt <prompt> to run and log.
   • Web App (web.py/api.py):
     • POST /api/generate to run prompts.
     • GET /api/usage/all / model/<model> / timeframe to retrieve logs.
   • Setup Script (setup.py): Initializes the DB and rebinds Ollama port.

File Structure

greenprompt/
├── cli.py                  # CLI entry point
├── core.py                 # Prompt runner & energy estimator
├── dbconn.py               # SQLite connection, init, and save/query functions
├── samplerMac.py           # macOS PowerMonitor for sampling power
├── sysUsage.py             # OS-agnostic wrappers & estimators
├── setup.py                # Tool setup: DB init, Ollama rebind
├── analytics.py            # Plotly graphs for the dashboard
├── web.py                  # Flask web server, HTML dashboard
├── templates/
│   └── index.html          # Web UI template
├── static/                 # Optional assets (CSS)
└── constants.py            # User-editable settings

Installation & Setup

PyPI Installation

You can install GreenPrompt directly from PyPI:

pip install greenprompt

Or, can build it using Poetry

# If you don't have Poetry, install it first:
pip install poetry

# Install dependencies and build the package
poetry install

# (Optional) To build a distributable wheel and sdist:
poetry build

And install it

# If you want to install the built package locally (after poetry build):
pip install dist/*.whl

Or, if you just want to install in your current environment for development:

poetry install

This will install both the greenprompt and gp CLI commands.

Development Installation

2. Clone & Install

   git clone <repo-url>
   cd greenprompt
   poetry install
   

3. Initialize

   python setup.py
   

   • Creates greenprompt_usage.db
   • Rebinds Ollama to port 11435 (via setup instructions)

Usage

CLI Usage

Once installed, use the following commands:

# Setup Ollama port, initialize database, and generate constants
greenprompt setup
# or using alias
gp setup

# Start the web API server and dashboard
greenprompt run --port 5000
gp run --port 5000

# Send a prompt and display stats
greenprompt prompt "Hello, world!" --model llama2
gp prompt "Hello, world!" --model llama2

# Monitor the last 10 prompt usage entries
greenprompt monitor --count 10
gp monitor --count 10

# Launch Dashboard
greenprompt dashboard
gp dashboard

API Endpoints

• POST /api/prompt
  Send a JSON body with prompt and optional model.
  Returns the generated response and all energy/token metrics.

• GET /api/usage/all
  Retrieve all prompt usage records as a JSON array.

• GET /api/usage/model/
  Retrieve prompt usage records filtered by model.

• GET /api/usage/timeframe?start=ISO&end=ISO
  Retrieve prompt usage records between start and end ISO timestamps.

• GET /dashboard
  Serve the HTML dashboard with embedded Plotly analytics.

• Proxy all Ollama API calls via ANY /ollama/api/
  Forward requests to the local Ollama server (default port 11434), preserving method, headers, query params, and body.

Limitations

• macOS only (relies on powermetrics)
• Per-process power usage is estimated proportionally from system metrics
• CLI proxy requires redirecting Ollama to use 127.0.0.1:5000:11434

Future Work

• Linux & Windows support (via Intel Power Gadget, rapl, nvidia-smi)
• VS Code & browser extensions
• Carbon offset integrations
• Team dashboards & enterprise reporting

License

MIT License