oagi-lux-samples
Public
Pull mirror status: ok
T
chore: use 10 as scroll amount for macos
tasker_examples
chore: use 10 as scroll amount for macos
.gitignore
Begin removing the gemini VLM for dedicated examples
README.md
Create examples without VLM analysis

OAGI Lux Samples

This repository demonstrates how to leverage Lux in various use cases. Lux is a Computer Use foundation model from OpenAGI that enables you to build AI agents capable of automating desktop and web tasks through visual understanding and action execution.

What is Lux?

Lux provides a TaskerAgent that can:

  • Execute multi-step workflows defined as "todos"
  • Take screenshots and understand UI elements visually
  • Perform actions like clicking, typing, and scrolling via PyAutoGUI
  • Track execution history and export detailed reports

Getting Started

  1. Install the oagi package: 
    pip install -r tasker_examples/requirements.txt
  2. Set your API key: 
    export OAGI_API_KEY="your-api-key"

Examples

All examples are located in the tasker_examples/ directory.

Amazon Scraping

Location: tasker_examples/amazon_scraping/

Searches Amazon for a product and sorts results by best sellers.

python tasker_examples/amazon_scraping/amazon_scraping.py --product_name "headphones"

Options:

  • --product_name - Product to search for (default: purse)
  • --exp_name - Experiment name for saving results (default: amazon_crawl)
  • --save_dir - Directory to save results (default: results/)
  • --model_name - Model to use (default: lux-actor-1)
  • --max_steps - Max steps per todo (default: 24)

CVS Appointment Booking

Location: tasker_examples/cvs_appointment_booking/

Navigates CVS.com to schedule a flu shot appointment by filling out forms and selecting options.

python tasker_examples/cvs_appointment_booking/cvs_tasker.py \
    --first_name "John" \
    --last_name "Doe" \
    --email "john@example.com" \
    --birthday "01-15-1990" \
    --zip_code "10001"

Options:

  • --first_name - First name (default: First)
  • --last_name - Last name (default: Last)
  • --email - Email address (default: user@example.com)
  • --birthday - Birthday in MM-DD-YYYY format (default: MM-DD-YYYY)
  • --zip_code - ZIP code for location search (default: 00000)
  • --exp_name - Experiment name (default: cvs)
  • --save_dir - Directory to save results (default: results)

Software QA with Nuclear Player

Location: tasker_examples/software_qa_with_nuclear/

Automates UI testing of the Nuclear Player app by clicking through all sidebar buttons and verifying each page loads correctly.

python tasker_examples/software_qa_with_nuclear/software_qa.py

Options:

  • --exp_name - Experiment name (default: nuclear_qa)
  • --save_dir - Directory to save results (default: results/)
  • --model_name - Model to use (default: lux-actor-1)
  • --max_steps - Max steps per todo (default: 24)

Note: This example requires the Nuclear Player app to be installed and running.

Key Components

  • TaskerAgent - The core agent that executes todo-based workflows
  • AsyncScreenshotMaker - Captures screenshots for visual analysis
  • AsyncPyautoguiActionHandler - Executes mouse/keyboard actions
  • AsyncAgentObserver - Records execution history for debugging