wjhccc/tradingagents-studio

TradingAgents-Studio is a research workbench that lets you watch a team of AI agents analyze a stock, debate the bull and bear cases, and arrive at a trading decision, all through a web interface. It is a fork of an earlier command-line framework called TradingAgents, extended with a visual layer and support for Chinese A-share markets. The README is explicit that this is a research and educational tool only, not investment advice. The core idea is transparency. Most AI trading tools hand you a final buy or sell call. This one shows you the reasoning process: a chain of cards tracing how a news event flows through supply chains and into sector sentiment, and a side-by-side chat-bubble format that separates the bullish and bearish arguments round by round. The debate view updates in real time over a live connection as the analysis runs, rather than appearing all at once at the end. The project has native support for Chinese market data. It auto-detects Chinese ticker formats and routes them through data sources including AKShare (free, the default) and optionally Tushare Pro. Four analysts specific to Chinese markets cover retail investor discussion forums, event-driven causal reasoning, capital flow data (including northbound stock connect flows and margin activity), and macroeconomic indicators like CPI, M2, and the LPR policy rate. Beyond single analyses, the tool includes several workflow features. You can track a portfolio of holdings with real-time quotes and see the latest signal for each position. A paper trading account lets you simulate trades based on agent decisions with one click, including enforcement of the Chinese T+1 settlement rule. A decision replay backtest replays previously stored agent calls over any historical window to show what the return, drawdown, and Sharpe ratio would have been, at no additional AI cost since it uses stored results. Analyses can also be scheduled to run automatically on an interval or daily basis, with automatic disabling after three consecutive failures. The backend is Python with FastAPI and the frontend uses Vue 3. It supports a wide range of AI providers including DeepSeek, Alibaba Tongyi, Zhipu, and Ollama for local models, alongside the major international providers. API keys can be managed from the settings page in the web UI and are written through to the environment file so the command-line version shares the same configuration.