What is Proxy Lite
Proxy Lite is an open-source lightweight vision-language model (VLM) with 3B parameters, supporting automated web tasks. Proxy Lite can operate a browser like a human, completing repetitive work such as web interaction, data scraping, and form filling, significantly reducing automation costs. Proxy Lite uses a three-step decision-making mechanism of "observe-think-tool invocation", with strong generalization ability and low resource usage, and supports efficient operation on consumer-grade GPUs. Proxy Lite provides a complete browser interaction framework, supporting various application scenarios such as web automation testing, data extraction, and intelligent task assistants, and is suitable for developers to quickly deploy and use.
Main features of Proxy Lite
- Web automation operations: Automatically complete various operations in the browser, such as clicking buttons, filling out forms, scrolling pages, and handling pop-ups.
- Web data scraping: Parse web page content and extract structured data, supporting data scraping from multiple types of web pages such as news, social media, and e-commerce platforms.
- AI-driven interactive web agent: Combine visual perception capabilities to implement UI automation testing and front-end behavior analysis.
- Intelligent task assistant: Help users search, filter, and summarize information, improving the web navigation experience.
- Low resource usage: Only 3B parameters, supports efficient operation on consumer-grade GPUs, with no need to rely on large-scale cloud computing resources.
Technical principles of Proxy Lite
- Vision-language model (VLM): Combine visual perception (image recognition) and natural language processing capabilities to understand and operate web page content.
- Three-step decision-making mechanism：
- Observe: Evaluate whether the previous operation was successful and obtain the current web page status information.
- Think: Reason based on the current page status and decide the next operation.
- Tool invocation: Interact based on browser APIs, such as clicking, inputting, and scrolling.
- Browser interaction framework: Built-in complete browser interaction control framework, using the Playwright library to drive browser operations, supporting Headless mode and Stealth mode to reduce the risk of anti-scraping detection.
- Execution feedback mechanism: Draw onDeepSeek R1The feedback mechanism of models such as these continuously optimizes the decision-making process during task execution and improves execution accuracy.
Proxy Lite project address
- GitHub repository：https://github.com/convergence-ai/proxy-lite
Application scenarios of Proxy Lite
- Web automation operations: Automatically complete tasks such as clicking, filling out forms, and scrolling pages, reducing manual operations.
- Web data scraping: Extract structured data from web pages such as news and e-commerce sites, for use in data analysis or content aggregation.
- Automated testing: Implement UI automation testing for web applications and quickly detect interface functions.
- Intelligent task assistant: Help users search and filter information, improving web usage efficiency.
- Enterprise-level task automation: Automate internal enterprise processes, such as data entry and publishing information across systems.