AutoPage is an open-source project released by the AutoLab-SAI-SJTU team at Shanghai Jiao Tong University. The paper has been accepted to ACL 2026 Findings, and the official implementation code is hosted on GitHub. The project aims to solve two problems: how to create webpages from academic papers, and how to evaluate the quality of project webpages.
In terms of use, AutoPage applies a human-machine collaborative multi-agent workflow to convert PDF papers into polished, directly publishable project webpages (HTML). At the same time, the project proposes PageBench to automatically evaluate generated webpages from two dimensions: content quality and visual design quality.
In terms of main capabilities, AutoPage supports flexible combinations of LLM / VLM, supports the use of models such as GPT-4o-mini and gemini-2.5-flash, and also supports deploying open-source models through vLLM. It can use templates in the templates folder and can also customize template styles (such as background_color, has_navigation, has_hero_section, title_color, page_density, image_layout). The project provides a Gradio frontend and also offers a command-line quick start method, and includes a complete Gradio demo video to explain the usage workflow and interface operations. During evaluation, the generated project pages are stored in the generated_project_pages directory. AutoPage is built on CAMEL. If you need to add a model, you can add the configuration in get_agent_config() in utils/wei_utils.py, and add the corresponding ModelType in camel/types/enums.py.
In terms of usage, installation requires installing Playwright; create a .env file in the root directory and add API keys (such as GEMINI_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY); create a folder named pdfs under AutoPage and place papers in it named as paper_name.pdf; when using open-source models, you need to first deploy with vLLM and correctly specify the port in the get_agent_config() function in utils/wei_utils.py. You can run the frontend through Gradio. If you choose a template in the templates folder, you need to fill in template_dir and template_file. The documentation gives examples of templates/airi-institute.github.io and templates/airi-institute.github.io/HairFastGAN/index.html; if you customize the style, leave template_dir and template_file empty and set the relevant style parameters.
In terms of applicable scenarios, it is suitable for researchers and developers who want to quickly convert academic papers into project homepages, and is also suitable for research work interested in paper-to-webpage generation and automatic evaluation. Related projects also include Paper2Poster, P2P, Paper2Video, Preacher, PresentAgent, AutoPR, PPTAgent, Paper2Agent, etc. AutoPage is regarded as part of the Paper2Anything ecosystem.
In terms of limitations, the existing materials do not provide explanations of charges, pricing, platform compatibility, or specific performance metrics; some functional details (such as complete installation commands and code) need to be found in the project repository; open-source models need to be deployed by yourself; and using commercial models requires configuring the corresponding API key yourself.
