Amazon Nova Act – A general-purpose AI agent launched by Amazon, capable of autonomously performing web tasks

What is Amazon Nova Act?

Amazon Nova Act is a general-purpose AI agent launched by Amazon AGI Labs, designed to perform tasks within web browsers. Amazon Nova Act supports developers in building intelligent agent application prototypes based on its accompanying SDK, enabling operations such as submitting leave applications, scheduling calendar events, or sending automated reply emails. By decomposing complex tasks into atomic commands and leveraging Playwright to directly interact with browsers, Amazon Nova Act enhances the reliability and efficiency of task execution. Amazon Nova Act empowers developers to create intelligent agent applications capable of automating web operations, with the potential to handle more complex, multi-step tasks in the future.

The main functions of Amazon Nova Act

• Decompose Complex Tasks: Supports developers in breaking down complex, multi-step tasks into reliable atomic commands, such as searching, filling out forms, clicking buttons, etc.
• Add Detailed Instructions: Supports developers in adding more detailed instructions for each command, such as “Do not accept insurance sales pitches” or “Select a specific date”, ensuring that tasks are executed as expected.
• Support Multi-task Operations: Capable of handling multiple tasks simultaneously. Tasks are parallelized and processed based on Python code to improve efficiency.
• Provide SDK Toolkit: Amazon has launched a supporting software development kit (SDK) to help developers build prototypes of intelligent agent applications.

The technical principles of Amazon Nova Act

• Natural Language Processing (NLP): Based on pre-trained language models, it parses users’ intentions and converts them into specific task operations. NLP technology supports multilingual processing, enabling Amazon Nova Act to serve users in different languages globally.
• Web Automation Technology: Simulates human web browsing behavior using automation tools such as Selenium or Puppeteer. By parsing the HTML structure of web pages, it identifies form fields and automatically fills in user-provided data. It also supports extracting specific information from web pages, such as restaurant reservation confirmation numbers, order statuses, etc.
• Machine Learning and Reinforcement Learning: Machine learning algorithms optimize the efficiency and accuracy of task execution. Based on extensive training data, the model learns the optimal task execution paths. Learning techniques enable Nova Act to self-optimize based on feedback during task execution. For example, if an error occurs while booking a restaurant, reinforcement learning adjusts the strategy to avoid making the same mistake in the future.
• API Calls and Integration: Interacts with various external services (such as restaurant reservation systems, online shopping platforms, etc.) through API calls to complete complex tasks. Amazon Nova Act provides an SDK, allowing developers to combine their own business logic to develop more customized applications.

The project address of Amazon Nova Act

• Project official website: https://labs.amazon.science/blog/nova-act

Application scenarios of Amazon Nova Act

• Office Automation: Automatically submit leave applications, arrange meetings, fill out forms, and generate reports.
• E-commerce: Automatically complete online shopping, regular procurement, and price monitoring.
• Personal Life Management: Automatically manage schedules, arrange a healthy lifestyle, and book meals.
• Complex Task Handling: Automatically organize events (such as weddings, parties) and execute complex IT tasks.
• Corporate Business Processes: Automatically handle customer relationships, supply chain management, and data analysis.