AI Browser
The AI Browser lets Halo's Agent directly interact with web pages — automatically opening pages, searching for information, filling out forms, and clicking buttons, just like an automated assistant that can see, click, and type.
How to Enable
Click the 🌐 icon at the bottom-left of the input box. When the icon is highlighted, the AI browser is active.
Once enabled, the AI will automatically decide during the conversation whether it needs to open a browser — no manual triggering required each time.
What the AI Can Do
- Automatically open web pages and navigate to specified URLs
- Take screenshots to recognize page content (requires a multimodal model such as Claude 4.5)
- Automatically fill out forms and click buttons
- Extract page information and return it to you in a structured format
Human-AI Collaboration
When the AI is operating the browser, the interface displays an "AI is operating this browser" notice along with a Live indicator.
You can also take over at any time: type an address directly in the browser or click to navigate — you and the AI can operate simultaneously without conflict.
Typical Use Cases
Online Shopping
Search for mechanical keyboards on Amazon, filter for models priced between $50–$120 with the highest ratings, and tell me the pros and cons of each.Information Gathering
Open Hacker News, summarize the 5 most-commented stories today, and give me a digest.Form Auto-Fill
Open the internal ticketing system and fill in today's work log: completed unit tests for the user login module.Notes
Multimodal model required
Page screenshot recognition requires a vision-capable model, such as Claude 4.5 or GPT-4o. When using a text-only model, the AI can only understand the page through its DOM structure and cannot recognize image content.
This feature is still being actively improved
The AI browser may be unstable in complex interaction scenarios. If you encounter issues, please report them on GitHub Issues.