Vibe-Trading: research agents and setup overview

Vibe-Trading is presented as an open source research workspace that turns natural language prompts into market analysis workflows. This narrated repository overview covers data loading, strategy generation, backtests and reports, plus persistent research memory. An AI agent can revise trading ideas, while teams of agents debate strategies and test them. Shadow accounting compares real trading records with a rule-based replica of a strategy.

The described setup starts with a Python package and a .env file containing an LLM provider and key. The speaker describes an interactive terminal and a web UI served by FastAPI locally. Example tasks include testing a moving average strategy for 2024 and summarizing return and drawdown. The source describes remote LLM provider choices; it does not establish offline operation or local model inference. The example prompts are explained rather than executed with a verified backtest result.

The overview also covers paper trading through local TWS or IB Gateway, optional live connectors, and exports to TradingView, TDX or MetaTrader. API and MCP interfaces connect the workspace to other services. The speaker says the development server binds to localhost and remote endpoints need an API authentication key. Docker and local installation are discussed, but no hardware requirements are given.

The closing limitation matters: the speaker describes Vibe-Trading as MIT-licensed research software that holds no funds and operates no execution venue. Broker trading is experimental and subject to user-defined limits.