BettaFish is a multi-agent public opinion analysis system built from scratch without framework dependencies. It enables users to analyze sentiment across 30+ domestic and international social media platforms and millions of user comments through natural conversation. The system features AI-powered full-domain monitoring with 24/7 crawler clusters tracking 10+ social platforms, composite analysis beyond single LLM approaches using five specialized agent types combined with fine-tuned models and statistical methods, and multimodal capabilities processing short-form video content and extracting structured information from search results. Its architecture consists of five primary agents: Query Agent for news breadth search, Media Agent for multimodal content understanding, Insight Agent for private database mining, Report Agent for intelligent multi-round report generation, and Forum Engine for agent collaboration through structured dialogue. A unique forum collaboration mechanism allows agents with distinct tool sets to engage in structured debate guided by an LLM moderator, preventing homogenization and producing higher-quality collective intelligence. Supporting systems include MindSpider for social media crawling, SentimentAnalysisModel for multiple sentiment analysis implementations, and ReportEngine for GraphRAG knowledge graph construction with rendering to HTML/PDF. The project is deployable via Docker Compose or source code setup requiring Python 3.9+.