bettafish

Micro Opinion (BettaFish) is an innovative multi-agent public opinion analysis system built from scratch without relying on any framework. Users only need to raise analysis requests as if chatting, and the system can fully automatically analyze 30+ mainstream social media platforms at home and abroad and millions of public comments, generating in-depth research reports. The system core consists of four engines: Query Agent is responsible for broad news search at home and abroad; Media Agent has powerful multimodal content analysis capabilities (can parse short videos, images, and structured information cards from search engines); Insight Agent performs deep mining on private databases; Report Agent uses built-in templates for multi-round report generation. In addition, the ForumEngine forum engine gives different agents unique tool sets and thinking modes, introduces a debate moderator model, and through chain-of-thought collisions and debate avoids the limitations of a single model's thinking, fostering higher-quality collective intelligence. In terms of technical implementation, the system adopts a pure Python modular design, supports PostgreSQL and MySQL databases, and all LLM calls are compatible with OpenAI API standards, allowing integration with any model provider that conforms to the standard. The sentiment analysis module integrates multiple methods, including fine-tuned BERT/GPT-2 models, multilingual sentiment analysis, small-parameter Qwen3 fine-tuning, and traditional machine learning methods, which can be flexibly switched according to needs. In terms of deployment, the system provides a one-click Docker deployment solution and also supports source code startup. The crawler system MindSpider can run independently, supporting topic extraction and deep public opinion crawling. Report generation supports three output formats: HTML, PDF, and Markdown, and provides a command-line tool for quick retry and debugging. The system also supports integration with custom business databases, breaking down data barriers between public and private domains, and providing vertical businesses with a fusion analysis capability of external trends and internal insights. The project code structure is clear. Each Agent engine contains an independent LLM interface wrapper, processing nodes, tool sets, state management, and prompt templates, allowing developers to easily integrate custom models and business logic. The system also provides a Streamlit single-Agent application for independent debugging of each engine. The overall design emphasizes lightweight and high extensibility, aiming not only at public opinion analysis but also at becoming a simple and universal data analysis engine driving all business scenarios.