What is this?
BettaFish (Chinese name 'Weiyu'/微舆) is a from-scratch multi-agent public opinion analysis system that monitors social media and large volumes of public comments, then produces reports to help break information bubbles and support decision-making.
BettaFish (Chinese name 'Weiyu'/微舆) is a from-scratch multi-agent public opinion analysis system that monitors social media and large volumes of public comments, then produces reports to help break information bubbles and support decision-making.Is this built on top of an agent framework like LangChain?No — it is implemented from scratch in pure Python with no dependency on any agent framework. It is a lightweight, modular, self-hosted system of its own Query/Media/Insight/Report agents plus a Forum Engine for coordination.
No — it is implemented from scratch in pure Python with no dependency on any agent framework. It is a lightweight, modular, self-hosted system of its own Query/Media/Insight/Report agents plus a Forum Engine for coordination.How do I get it running?It ships with Docker and Docker Compose configs for one-click deployment, or you can set up a Python environment and run the Flask main app directly; each agent engine (Query/Media/Insight) also has a standalone Streamlit app you can run on its own.
It ships with Docker and Docker Compose configs for one-click deployment, or you can set up a Python environment and run the Flask main app directly; each agent engine (Query/Media/Insight) also has a standalone Streamlit app you can run on its own.Do I need an OpenAI-compatible LLM API key to run the agents?Yes — every agent engine (Query, Media, Insight, Report) talks to models through a unified OpenAI-compatible client, so you need to provide API credentials for whatever LLM provider you configure.
Yes — every agent engine (Query, Media, Insight, Report) talks to models through a unified OpenAI-compatible client, so you need to provide API credentials for whatever LLM provider you configure.Which social platforms does the built-in crawler actually cover?The crawler cluster is built around major Chinese social platforms — Weibo, Xiaohongshu, Douyin, Kuaishou and other domestic/international networks — so out-of-the-box coverage outside these ecosystems is limited.
The crawler cluster is built around major Chinese social platforms — Weibo, Xiaohongshu, Douyin, Kuaishou and other domestic/international networks — so out-of-the-box coverage outside these ecosystems is limited.What is the Forum Engine and why does it matter?It lets the Query, Media and Insight agents debate under an LLM moderator inside a shared 'forum', avoiding the groupthink of a single model and producing higher-quality collective analysis.
It lets the Query, Media and Insight agents debate under an LLM moderator inside a shared 'forum', avoiding the groupthink of a single model and producing higher-quality collective analysis.What sentiment-analysis models does it ship with?The SentimentAnalysisModel folder includes a LoRA-fine-tuned Chinese BERT model, a GPT-2 LoRA model, a small fine-tuned Qwen3 model, a multilingual sentiment model, and classic machine-learning baselines.
The SentimentAnalysisModel folder includes a LoRA-fine-tuned Chinese BERT model, a GPT-2 LoRA model, a small fine-tuned Qwen3 model, a multilingual sentiment model, and classic machine-learning baselines.Can I repurpose BettaFish for a different domain, like finance?Yes — the author notes you can turn it into, for example, a financial market analysis system just by changing the API parameters and prompts used by the agents' tool sets.
Yes — the author notes you can turn it into, for example, a financial market analysis system just by changing the API parameters and prompts used by the agents' tool sets.What format are the final reports in?Reports are assembled from a structured intermediate representation (IR) and rendered into interactive HTML, with PDF export available via WeasyPrint.
Reports are assembled from a structured intermediate representation (IR) and rendered into interactive HTML, with PDF export available via WeasyPrint.What are similar projects?DeepAnalyze, ChatDev, PaddleNLP, jarviscore-framework, MediaCrawler, catbus, MiroFish, Transformers-Tutorials
DeepAnalyze, ChatDev, PaddleNLP, jarviscore-framework, MediaCrawler, catbus, MiroFish, Transformers-TutorialsWhich topics is it in?gpt-2, weibo, bert, sentiment-analysis, streamlit, douyin, deep-research, xiaohongshu, flask, crawler, llms, multi-agent-systems
gpt-2, weibo, bert, sentiment-analysis, streamlit, douyin, deep-research, xiaohongshu, flask, crawler, llms, multi-agent-systems