We have created a new AI module for Browser Automation Studio that can solve different types of CAPTCHAs using Astra, DeepSeek, and other models of your choice.

The module is simple to use: call a BAS action and provide the API keys required by the selected AI services. You do not need to specify a selector for the CAPTCHA element or build additional logic to find it on the page.

How it works

The task is automatically divided between two models:

1. A less expensive model, DeepSeek by default, analyzes the page, finds the CAPTCHA, crops the relevant area, and determines whether it needs to be solved.
2. A more powerful model receives the detected CAPTCHA and completes the actual challenge.

This approach can handle unusual and rare CAPTCHA types, including challenges that require dragging elements or completing several actions in sequence.

The module solves CAPTCHAs through regular browser actions such as clicks and drag-and-drop. It does not inject code into the page, making this interaction more difficult to distinguish from that of a regular user.

Cost and future improvements

At the moment, this solution costs more than specialized CAPTCHA-solving services. Complex CAPTCHA types may cost around $0.01 per solution. We expect the cost to decrease as more capable and affordable AI models become available.

Demonstration

The video shows BAS solving several unusual and fairly complex CAPTCHA types with AI:

This is currently a demonstration of the module. It has not yet been added to a BAS build.