Amazon.com Inc. today took another step into agentic artificial intelligence with the general availability of Nova Act, a browser-controlling AI system that can independently navigate websites.

Initially unveiled in March, Nova Act provides an agentic foundation that allows AI systems to complete tasks in web browsers on behalf of the user: clicking buttons, filling text fields and navigating sites. This enables the agent to browse the web, interact with websites and complete multistep actions end-to-end.

Amazon announced the new AI agent during AWS re:Invent 2025. It was developed by Amazon AGI Lab, which is a part of Amazon’s AGI organization that released the Amazon Nova foundation models last year. Nova Act is powered by a custom Nova 2 Lite model, a small and fast multimodal model in the Amazon Nova family, and trained specifically for this purpose.

Browser-capable AI agents can extend consumer capabilities by entering ecommerce sites, searching for products and completing purchases. They can quickly compare products, services and vendors, then make intelligent decisions about what to buy. For businesses, they can read sites for data, chain multiple steps across different domains to finish complex workflows and automate repetitive tasks across teams.

Organizations can also use Nova Act to build fleets of enterprise-ready, browser-capable agents capable of handling the same sort of visual “thinking” humans use when interpreting web pages.

“We believe [the new model] will be as good, if not better, than any existing models in the market — and more cost-effective,” Elan Dekel, head of Nova Act product at Amazon AGI Lab, told SiliconANGLE in an interview.

Alongside the general availability release, Amazon introduced a new no-code playground designed for users of all technical skill levels to prototype and design agents.

“We want to make it very easy for less techy people… maybe a company executive who hasn’t coded in a few years,” Dekel said.

Other new tools include a Visual Studio Code extension with a notebook-style user experience, an embedded browser and fully integrated deployment workflow. Nova Act agents also support human-in-the-loop oversight, allowing teams to define when an agent should pause, request permission or check in for additional instructions. A new Nova Act Console will enable developers and operators to constantly monitor workflow runs and adjust them as needed.

Amazon also previewed new advanced capabilities such as tool calling, which will allow Nova Act to determine when it needs external software tools and when to use them. This extends Act’s capabilities beyond using the web browser and generating text to query databases, access real-time data, send emails, tweets and perform other actions outside of or to support actions.

Dekel explained that Amazon’s Nova Act is trained to understand complete workflows from inception to execution. Instead of stitching together separate components, the system natively grasps both how to act within browser-based environments and which tools to invoke in each context.

“You don’t have to mix and match tools from different providers… because it’s all trained together, the quality and reliability is higher,” Dekel said.

The general availability of Act comes at a time that web browsing capabilities are being explored broadly by major AI companies including OpenAI Group PBC, Google LLC and Anthropic PBC. Anthropic unveiled a version of its AI Claude that can use computer interfaces last year, including web browsers, and OpenAI released “ChatGPT Atlas,” a web browser with an embedded agent capable of performing actions on behalf the user.

Dekel explained that Nova Act will shift to an hourly-based pricing model once in general availability. This means that customers won’t be billed on a token basis or per API calls; instead, they’ll pay for the amount of time the agent is active, which more closely mirrors traditional hourly-work models. Ideally, this will make costs more predictable and transparent and also better support long-running automations.

Extended collaboration with 1Password

Amazon announced 1Password LLC as a launch partner for Nova Act, bringing 1Password’s credentials security management capabilities directly into agentic AI automation.

For the past year, 1Password and Amazon Web Services have collaborated to help customers adopt AI tools and streamline authentication for AI agents. The companies formalized this effort in June with a multi-year collaboration agreement.

“We want to extend our secure vaulting capabilities to every endpoint… not just devices, not just dev environments, but also the browser,” Nancy Wang, senior vice president of engineering at 1Password, told SiliconANGLE in an interview. “The definition of who counts as an AI developer has really broadened.”

As browser-capable AI agents become commonplace, they will need to “remember” passwords, credentials and other sensitive information so they can log into sites and access application programming interfaces. Just like humans, they must protect these secrets to prevent accidental exposure or malicious interception.

“All this requires access to secure credentials,” Wang said, referring to autonomous actions by AI agents. “All of that can be fetched right from 1Password’s secure vault, as the agent is going through these steps right in a user directed workflow.”

Over the past year, 1Password and Amazon have worked together to build this secrets protection into AI-driven autonomous workflows. For example, the release of the 1Password model context protocol server for Trelica helps information technology teams identify unmanaged apps, reduce AI “shadow IT” and provide identity-based access controls for AI agents. AWS Secrets Sync connects 1Password directly into AWS Secrets Manager, allowing a developers, IT admins and security teams to maintain passwords and secrets in one place and have them securely synchronize.

To protect AI actions and secrets, 1Password relies on AWS Nitro Enclaves, the secure, isolated and hardened environments within Elastic Compute Cloud instances. This architecture extends encryption from source to cloud, ensuring sensitive data is processed inside attested virtual machines — meaning secrets are never exposed to outside parties, including 1Password itself.

Image: SiliconANGLE/Microsoft Designer