What it does

BettaFish (微舆) is a from-scratch Python system that turns a chat-style question into a full sentiment-analysis report. It deploys crawler agents to monitor 30+ Chinese and international social platforms, then runs a multi-agent “forum” where specialized LLMs debate findings before a Report Agent renders the output as interactive HTML.

The interesting bit

The project avoids every major agent framework and builds its own orchestration. The ForumEngine acts like a debate moderator: agents with distinct toolkits and personas research in parallel, then a host LLM challenges their conclusions, forcing course corrections across multiple rounds. It’s a deliberate attempt to break “information cocoons” by making models disagree with each other.

Key highlights

• Five specialized engines: Query (web search), Media (multimodal/video parsing), Insight (private DB mining), Forum (debate coordination), and Report (templated HTML generation)
• Pure Python modular design with explicit nodes, state management, and prompt templates per engine
• Claims 7×24 AI crawler coverage of Weibo, Xiaohongshu, Douyin, Kuaishou, and others
• Supports private database integration via SQLAlchemy async read-only queries
• Docker-ready with one-click deployment claims; version 1.2.1 as of README
• Companion project MiroFish extends the pipeline into predictive crowd-intelligence

Caveats

• README is heavy on marketing language (“break information cocoons,” “restore truth”) and light on reproducible setup details
• Actual crawler implementation and rate-limiting behavior are not shown in the truncated source
• Sponsor integrations (AIHubMix, Anspire) are prominently featured; some core search capabilities may depend on these paid APIs

Verdict

Worth a look if you’re building Chinese-market sentiment tools or studying multi-agent consensus patterns without LangChain cruft. Skip if you need battle-tested, documented infrastructure—this is ambitious research code with 41k stars but thin operational guidance.