AI BriefingWebBrain releases open-source browser agent with 450M local vision model
Key point
The project introduces webbrain-vl-2-450M, a specialized vision model trained on browser screenshots that runs locally via WebGPU.
Details
WebBrain is an open-source browser agent designed to operate independently of cloud providers by supporting local models through LM Studio and Ollama. The project introduces webbrain-vl-2-450M, a vision-language model fine-tuned on browser screenshots and tasks based on LFM-2.5-VL-450M. This small model enables local browser perception via WebGPU, reducing the need for large multimodal models for every screenshot.
Architecture and Capabilities
The agent combines screenshots with the browser accessibility tree for perception, avoiding reliance on pure DOM parsing. Key features include:
- Model-agnostic planner: Supports various local or cloud LLMs.
- Browser support: Compatible with Chrome, Edge, Firefox, and Chromium.
- Local execution: The 450M VLM runs directly on the user's machine.
- Open source: Code, model, and dataset are publicly available.
The project aims to demonstrate the viability of browser agents using small local models rather than GPT or Claude-scale APIs.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Ollaya Launches Local Runtime for Open-Source Decision Models with TypeSafe SDK Compatibility
Whiteboard Launches Open-Source VS Code Fork for Human-AI Collaborative Software DesignOpenComputer Hosts 'Ship Video' Agent Using Claude Opus 5.5 to Generate Explainer Videos from Code