BettaFish is a multi-agent public opinion analysis system built from scratch in Python without relying on external frameworks. The project, whose name references the betta fish symbolizing small yet powerful and fearless characteristics, aims to help users break through information bubbles, restore the true nature of public sentiment, predict future trends, and support decision-making. Users interact with the system conversationally, submitting analysis requests that trigger automated analysis across more than 30 mainstream domestic and international social media platforms and millions of public comments.
The system architecture comprises four specialized agents working in coordinated fashion. The Query Agent provides precise information search capabilities with domestic and international web search abilities. The Media Agent handles multimodal content analysis with strong capabilities for processing diverse content types. The Insight Agent performs private database mining and deep analysis of proprietary sentiment data. The Report Agent generates intelligent reports using built-in templates. These agents operate through a forum-style collaboration mechanism where different agents possess unique tool sets and thinking modes, with a debate moderator model facilitating chain-of-thought collisions and debates to avoid homogenization and generate higher-quality collective intelligence.
The analysis workflow follows a structured process beginning with user queries received by the Flask main application, followed by parallel agent activation. After initial analysis and strategy formulation, the system enters a multi-round loop phase where agents conduct deep research guided by the forum moderator while the ForumEngine monitors agent contributions and generates moderator guidance. Results are then consolidated by the Report Agent, converted to an intermediate representation with dynamically selected templates and styles, and finally rendered into interactive HTML reports with quality checks applied during block-by-block generation.
According to GitGenius tracking data, the repository shows strong community engagement with 428 tracked issues and pull requests. The median issue and PR response latency is 0.1 hours with a mean of 48.1 hours, indicating active maintenance. The most frequently discussed topics center on LLM API integration with 123 labeled items, followed by Q&A discussions with 48 items and improvement suggestions with 25 items. Primary contributor 666ghj has logged 421 events, with DoiiarX contributing 376 events and NTFago adding 39 events. The project's contributor base overlaps with major repositories including microsoft/vscode, microsoft/typescript, and rust-lang/rust, suggesting involvement from experienced open-source developers.
The system distinguishes itself through six key advantages: AI-driven full-domain monitoring with crawler clusters operating continuously across multiple social media platforms, a composite analysis engine combining five professional agent types with fine-tuned models and statistical models, strong multimodal capabilities extending to short video content analysis and structured information extraction from search engines, the forum-style agent collaboration mechanism, seamless fusion of public and private domain data through secure interfaces, and lightweight Python-based modular design enabling one-click deployment with clear code structure for easy customization. The project positions itself as beginning with public opinion analysis but extending toward becoming a general-purpose data analysis engine applicable across various business scenarios through simple modifications to agent tool sets and prompts.
