Backtest a trading strategy on historical stock price data by describing it in plain English to the AI agent.
Screen stocks based on financial metrics like return on equity or revenue growth without writing any code.
Track a paper portfolio alongside real prices to practice investment decisions without risking real money.
Run a Swarm multi-agent analysis on a stock to get perspectives from multiple AI workers synthesized into one report.
Requires a Tushare API key for Chinese A-share market data, US market data sources require separate configuration.
Vibe-Trading is an AI-powered trading assistant you install locally and talk to in plain language to research stocks, run backtests, and track a portfolio. Rather than a traditional trading platform with fixed screens and buttons, it presents itself as an agent you converse with: you describe what you want to analyze, and it fetches market data, runs calculations, and returns results. It does not place real orders, it is a research and analysis tool. The core workflow centers on a chat interface available both in a web browser and on the command line. From there you can ask the agent to backtest a trading strategy across historical price data, screen stocks based on financial metrics, produce correlation heatmaps, or analyze dividends. For Chinese A-share markets it integrates with the Tushare data provider, which supplies both price history and point-in-time fundamental data such as revenue and return on equity, allowing strategies to be tested without accidentally using financial figures that were not yet public on any given historical date. The project also supports a mode called Swarm, where multiple AI workers run in parallel to analyze a topic from different angles before synthesizing a combined report. A Shadow Account feature lets you track a paper portfolio alongside real prices without risking actual money. Persistent memory lets the agent remember notes and context across sessions. Installation is a single pip command. The backend is built with FastAPI and the web interface uses React. The agent can be connected to other tools via the Model Context Protocol, which is an emerging standard for wiring AI assistants to external data sources and capabilities. The project is actively developed and releases updates frequently. Documentation and the README are available in English, Chinese, Japanese, Korean, and Arabic. The full README is longer than what was shown.
← hkuds on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.