Juanji: An AI Agent Code Tool by the OpenBMB Open-Source Community 139

小编 · 2025-04-24

Juanji is a recently launched AI Agent code tool by the OpenBMB open-source community, designed to efficiently extract valuable content. Users simply enter their requirements on the official website and quickly receive a comprehensive overview report.

Juanji: An AI Agent Code Tool by the OpenBMB Open-Source Community

What is SurveyGO?

"SurveyGO" is a code AI Agent tool recently launched by the OpenBMB open-source community, specifically designed to efficiently extract valuable content. Users simply need to enter their requirements on the official website, and a comprehensive survey report can be quickly generated.

"SurveyGO" offers two modes to choose from: the Standard mode outputs titles and keywords; the Professional mode is more flexible, allowing users to customize the source of materials, such as choosing online retrieval or uploading files.

Compared to other tools, "SurveyGO" stands out for its rigorous logic and academic quality, making it particularly suitable for in-depth analysis. The articles it generates have a clear structure, with introductory sections that are easy to understand and thorough analyses at the end, featuring detailed and well-supported arguments.

The technology behind "SurveyGO" is LLMxMapReduce-V2, jointly developed by multiple teams and specifically designed to optimize literature aggregation and utilization. Currently, "SurveyGO" is live, and the related papers and open-source content are both available.

SurveyGO: AI Agent code tool launched by the OpenBMB open-source community.webp

Core Features of SurveyGO

Literature Filtering and Structured Output: Automatically filters high-quality literature from paper repositories based on user-input keywords, and generates a structured framework containing core logic, methodology, and research gaps.

Dynamic Knowledge Graph Construction: Analyzes citation relationships and conceptual connections among literature to generate a visualized evolution map of the research field, helping users quickly locate academic hotspots and trends.

Multilingual Support and Domain Adaptation: Supports mixed Chinese-English input, and can adjust the analysis model according to different disciplines such as computer science and biomedicine, enhancing domain adaptability.

Technical Principles of SurveyGO

"SurveyGO" is developed based on the OpenBMB model ecosystem, combining convolution algorithms with information entropy optimization techniques to rapidly parse massive literature and generate logically clear, high-information-density survey reports. It is powered by the LLMxMapReduce-V2 long-text integrated generation technology, jointly developed by AI9Star, OpenBMB, and the Tsinghua University team. The core lies in aggregating reference materials through text convolution algorithms, combined with an information entropy-driven random convolution test-time scaling method, to efficiently handle ultra-long inputs and improve article quality.

SurveyGO Tutorial

Open the website https://surveygo.thunlp.org/ and click [Start Survey].
Choose Standard mode or Professional mode to submit your requirements.

Standard mode: Output the title and keyword description, then submit and wait for generation.
Professional mode: You can further customize the source of materials, choosing "Online Retrieval" or "Upload Files". To protect user privacy, articles generated based on uploaded files and the uploaded files themselves can be set as non-public.

Visit the website again to retrieve the generated survey report.

SurveyGO Tutorial.webp

Performance

Speed and Quality: In test cases by the Tsinghua University team, "SurveyGO" completed a survey of 200 papers in the field of quantum computing in just 3 minutes, producing an output of 28,000 words with logical coherence recognized by the supervising group.
Comparative Advantage: The team conducted a benchmark comparison, having "SurveyGO" and models such as OpenAI-DeepResearch write a survey on "The Impact of the Tariff War on the Lives of Ordinary People". Evaluated from the dimensions of structure, content, viewpoint, and citations, "SurveyGO" performed exceptionally well. The table of contents in its generated article is clearly layered, the content analysis is in-depth, the viewpoints are well-supported, and citations are precise, with overall strength exceeding other models.