Most multi-agent trading frameworks are built around US tickers and English-only news feeds, and they quietly break the moment you point them at a Shanghai-listed stock. This fork keeps the original TradingAgents agent-debate structure (created by Tauric Research) but re-plumbs everything underneath it for Chinese markets and Chinese LLMs — the part that is actually hard to retrofit.
What Sets It Apart
- Data layer rebuilt for A-share, HK and US coverage through Tushare, AkShare and BaoStock with fallback, so a run doesn't die when one source rate-limits or goes down.
- Model layer swaps the OpenAI-only default for domestic providers — Qwen/Tongyi, DeepSeek, Alibaba Bailian and the AiHubMix aggregator — with persistent model selection, which matters when US-hosted APIs are unreachable or too expensive.
- Output layer turns a multi-agent research run into a deliverable: reports export to Markdown, Word and PDF rather than living in a terminal log.
- Infrastructure is built for self-hosting: Docker multi-arch images (amd64/arm64), MongoDB plus Redis, and a move away from the upstream Streamlit toward a FastAPI + Vue web stack.
Who It's For and Tradeoffs
Great fit if you analyze A-share or HK equities, need Chinese-language reports, or are blocked from US-hosted model APIs. Look elsewhere if you only trade US stocks with reliable OpenAI access — the upstream TradingAgents stays leaner and closer to the underlying research. And take it on its own terms: the maintainer frames it as an educational and research platform, not trading advice or a backtested strategy.
