Vibe-Trading

Overview

Vibe-Trading is an AI-powered research workspace from HKUDS (Hong Kong University Data Systems group) that lets traders and analysts conduct market research through natural-language interaction with an intelligent agent. Rather than requiring users to write quantitative code directly, the platform translates plain-language questions into strategy generation, backtesting, and research artifacts across six asset classes. It hit 24.4K GitHub stars with a sustained growth rate of roughly 800 stars per day, making it one of the fastest-growing open-source trading tools in 2026.

The project occupies a middle ground between pure research frameworks like ai-hedge-fund and production trading platforms. It supports optional live execution through authorized brokers like Robinhood, but emphasizes user control — no funds are held by the platform, trading limits are user-set, and execution can be halted instantly.

Architecture

Self-improving agent: The core agent maintains persistent memory across sessions, accumulating market knowledge and refining its analysis over time. It connects hypothesis generation to automated backtesting through a Research Autopilot pipeline.

Multi-agent teams: Beyond the single-agent mode, Vibe-Trading supports multi-agent configurations that model investment committees, quant desks, and risk analysis teams — each agent specializing in a different analytical lens.

Alpha factor library: The platform ships with 452 pre-built alpha factors organized across multiple zoos including Qlib factors, academic factors from published research, and custom factor templates. These serve as building blocks for strategy construction.

Shadow Account: A distinctive feature that ingests broker transaction journals and uses LLM analysis to extract the implicit trading rules behind a user’s historical decisions — surfacing patterns the trader may not have consciously articulated.

Use Cases

Retail traders use Vibe-Trading to backtest strategy ideas expressed in plain language without writing quantitative code. The cross-market coverage means a single workspace handles US equities, crypto, forex, and Asian markets.

Quantitative researchers leverage the 452-factor library and Research Autopilot to rapidly iterate on hypotheses, connecting factor screening to backtesting in an automated loop.

Trading teams deploy multi-agent configurations to simulate investment committee dynamics — different agents presenting bull/bear cases, risk assessments, and position sizing recommendations before a portfolio decision.

Self-improvement workflows use the Shadow Account feature to analyze past trading behavior, extract implicit rules, and identify patterns of success or failure across historical transactions.

Considerations

While Vibe-Trading supports optional autonomous execution, users should treat live trading features with appropriate caution. The platform is research-first — the autonomous trading capability connects to authorized brokers but the quality of trading decisions depends entirely on the strategies and constraints the user configures. The MIT license means no warranty or liability coverage for trading losses. The 19 free data sources provide broad coverage but may have latency or quality limitations compared to institutional data feeds. The platform requires Python 3.11+ and runs a FastAPI backend with a React 19 frontend.

Who It’s For

Vibe-Trading is designed for retail traders, quantitative researchers, and small trading teams who want AI-assisted market analysis without building infrastructure from scratch. It is particularly useful for traders who think in natural language rather than code, researchers who need rapid hypothesis-to-backtest iteration, and anyone interested in multi-agent approaches to financial decision-making. The cross-market coverage makes it relevant globally, and the messaging channel adapters (16 platforms including Telegram, Slack, and WeChat) support team-based research delivery workflows.