Vibe-Trading: multi-agent trading research overview

Vibe-Trading is presented as an open source trading research workspace that turns natural-language requests into strategy code, backtests and reports. The speaker describes specialized agent teams for investment debates, quantitative analysis and cryptocurrency research, with market data passing through a centralized loader.

The setup overview covers pip installation and configuring an LLM provider's API key. The speaker also names Ollama for local inference without an API key. That option concerns model output; the broader workflow uses external market data and broker connections, so the video does not establish that the whole system works offline. The described stack uses Python, a React 19 frontend and a FastAPI backend.

Research examples include a moving-average backtest, factor evaluation and analysis of uploaded broker exports or CSV files. The speaker describes confidence intervals, forward validation and checks intended to prevent future information from leaking into backtests. Account analysis compares recorded trades with a rules-based historical alternative.

The broker section describes read-only account access and paper-account order support for some connectors. Claimed controls include exposure limits, a file-system kill switch and an audit trail. The overview also covers scheduled research and MCP access for compatible clients. These are the speaker's feature claims, rather than independent evidence of trading performance.