Overview

Nova Act is an experimental SDK from Amazon that enables developers to build browser automation agents. It combines natural language instructions with direct browser manipulation capabilities.

Key Features

  1. Hybrid Automation Approach:

    • Combines natural language instructions (act() method) with direct Playwright browser control
    • Allows breaking complex workflows into smaller, more reliable steps

  2. Information Extraction:

    • Supports structured data extraction using Pydantic models
    • Includes convenience schemas like BOOL_SCHEMA for simple yes/no questions

  3. Parallel Execution:

    • Enables running multiple browser sessions concurrently using ThreadPoolExecutor
    • Useful for tasks like scraping multiple pages simultaneously

  4. Authentication Handling:

    • Supports persistent browser state through Chrome user data directories
    • Allows pre-authenticated sessions

  5. Sensitive Data Handling:

    • Recommends using Playwright APIs directly for password entry
    • Provides warnings about screenshot collection of sensitive information

Technical Implementation

  • Built on Playwright for browser automation
  • Requires Python 3.10+
  • Supports MacOS and Ubuntu
  • Uses temporary directories for isolated browser sessions by default

Best Practices

  1. Prompt Design:

    • Be prescriptive and specific in instructions
    • Break large tasks into smaller steps
    • Avoid high-level, vague prompts

  2. Error Handling:

    • Check matches_schema when using structured responses
    • Handle potential schema mismatches gracefully

  3. Performance:

    • First run requires Playwright browser installation (1-2 minutes)
    • Subsequent runs start quickly

Limitations

  • Currently doesn't support IPython
  • Cannot interact with non-browser applications
  • Struggles with hidden elements (mouseover menus)
  • Doesn't handle browser window dialogs
  • Early research preview with many expected limitations

Use Cases

The SDK demonstrates several practical applications:

  1. E-commerce workflows:

    • Product search and cart operations
    • Order history management

  2. Data Collection:

    • Scraping structured data from websites
    • Aggregating information from multiple sources

  3. Task Automation:

    • Food ordering
    • Travel booking
    • Research tasks

Security Considerations

  • API keys must be protected
  • Sensitive data should be entered via Playwright, not natural language prompts
  • Screenshots may capture sensitive information visible in the browser
  • Includes Acceptable Use Policy requirements

Comparison to Alternatives

Compared to other automation tools:

  • More structured than pure Playwright scripting
  • More controllable than end-to-end LLM automation
  • Combines benefits of programmatic control with natural language flexibility

Getting Started

Basic requirements:

  1. Python 3.10+
  2. MacOS or Ubuntu
  3. API key from nova.amazon.com/act

Installation:

pip install nova-act

The SDK particularly suited for developers who need to automate web-based workflows while maintaining control over the automation process. Its hybrid approach offers a balance between flexibility and reliability that could make it useful for prototyping and certain production use cases.

Would you like me to help you using the Nova Act SDK?