Run financial research, backtests, and paper portfolios via conversational agents
You can ask questions in plain language without coding knowledge to receive investment reports and strategy validation results. Vibe-Trading is a tool that converts financial queries into research reports, executable backtests, and paper portfolios. The recent v0.1.19 release adds the ability to reuse backtests from previous sessions and apply version control to research artifacts, allowing for revisions supported by evidence.
Changes in v0.1.19
The core of this update is the persistence of the research process. Research artifacts generated during a conversation are now stored by version, and backtest results from previous turns or sessions can be reused immediately. You can also modify research goals in place, making it easy to update existing conclusions when new evidence emerges. Optional features now include 5-minute paper simulation for A-shares (Chinese stocks), Gildata-based financial lookups, stock screening, and portfolio diagnostics. The previously used QVeris has been removed, and the GILDATA_TOKEN alone no longer activates any functionality.
Conversational Research Workflow
Users ask questions in plain English via web chat or the terminal. For example, you can input "Backtest the 20/50 moving average crossover strategy for BTC-USDT in 2024, compare it with a buy-and-hold strategy, and report the maximum drawdown." The agent takes this request, identifies the trading pair, period, and benchmark, and retrieves historical price data for that period. After verifying the data source and dates, it writes the strategy, executes the backtest, and saves the strategy code, trade logs, and equity curve to the results directory. Finally, it reviews returns, turnover, drawdown, and validation warnings alongside the buy-and-hold comparison.
Data Sources and Market Coverage
The tool supports 28 market data sources, and when source: "auto" is set, it infers the market from the symbol and attempts sources in the configured order. For A-shares and HK stocks, it tries Tencent first, falling back to other sources like Eastmoney if it fails. US stocks prioritize Yahoo, while crypto assets are queried in the order of OKX, Binance, and CCXT. Chinese futures can use AKShare or local bars, and global futures require local bars. Installing vibe-trading-ai[ashare] adds a BaoStock loader that fetches A-share daily data without a token.
Broker Connections and Live Trading
You can retrieve broker holdings using a read-only profile or execute paper orders in a sandbox. Some brokers support live orders within the permission scope set by the user, with funds remaining directly with the funding broker. It supports various brokers including alpaca, binance, etoro, futu, ibkr, okx, robinhood, tiger, upbit, and zerodha. Before enabling live trading, you must set symbol, order, and exposure limits, and an emergency stop switch allows you to halt real-time activity immediately. A table specifies the read, simulation, and live support status for each broker.
Pre-Usage Checks
Python 3.11 or higher is required, and after installing with pip install vibe-trading-ai, you must run vibe-trading init to configure the model provider. To use the web app, run the frontend and API together via Docker, and set the provider and API key in the .env file before starting the container. If considering live trading, you must verify which profile (read, simulation, or live) the specific broker supports. Gildata-related features are optional, and free data sources are designed not to fall back to Gildata. Version control for research artifacts and backtest reuse are supported only in the latest version, so it is advisable to check your update status.
