The best quant tools never made it to retail investors—until someone built one for free and put the code on GitHub.
The Summary
- A Chinese developer released TickFlow Stock Panel, a self-hosted quantitative trading workbench for China's A-share market that scans 5,000+ stocks in milliseconds using Polars
- The platform integrates LLM-driven strategy generation, four-dimensional AI stock analysis (technical, fundamental, financial, news), and real-time monitoring with voice alerts and messaging integration
- Built for individual retail traders, not institutions—explicitly rejects "AI stock tips" hype and commercial use, positioning as an open-source learning tool
The Signal
Quantitative trading platforms typically cost thousands per month and run in institutional clouds. TickFlow Stock Panel flips that model: zero maintenance, runs locally, and scans China's entire A-share market in milliseconds. The developer built 18 pre-configured strategies plus custom signal builders, backtesting with T+1 settlement rules, transaction costs, slippage, and stop-losses.
The AI integration isn't cosmetic. The system uses LLMs to generate trading strategies from natural language, analyze individual stocks across four dimensions (technical patterns, fundamentals, financials, news sentiment), and write post-market reviews of consecutive limit-up stock tiers. Voice alerts call out stock names and signals in real-time. Feishu (China's Slack equivalent) pushes notifications when conditions trigger.
"Self-hosted, zero-maintenance A-share workbench for individual retail traders and quant enthusiasts."
What makes this technically interesting: pluggable data sources. The architecture accepts third-party feeds via API, webhooks, CSV, or JSON. You're not locked into one vendor's market data. The factor pipeline—MA, EMA, MACD, RSI, KDJ, Bollinger, volume ratio—scans and writes enriched Parquet files once, then queries at millisecond speed.
The monitoring system tracks four categories:
- Strategy triggers across the market
- Individual stock signals matching user-defined rules
- Price thresholds with AND/OR logic
- Exchange-flagged anomalies (3/10/30-day deviation metrics)
Market environment classification runs six emotional cycle phases driven by consecutive limit-up stock ladder data, plus concept and sector momentum rankings. The consecutive limit-up ladder page shows which stocks are hitting daily max gains for multiple days straight, breaking down by concept and industry, tracking seal orders (buy orders at the limit price that prevent further trading).
The developer's disclaimer is unusually direct: this is not investment software, not a charting app, and definitely not competing with established platforms like Tongdaxin or Flush. No "AI stock picks," no "limit-up predictions." Email for questions. Stars appreciated. A footer note mentions an exchange group QR code.
The Implication
Retail traders in China now have institutional-grade quant infrastructure they can run on a laptop. The LLM layer isn't a gimmick—it's doing the grunt work of translating ideas into testable strategies and synthesizing multi-source analysis humans don't have time for. Watch for forks adapting this architecture to other markets, and for brokerages to either embrace similar open tooling or double down on walled gardens.
If you're building AI agents for financial analysis, study this. The value isn't prediction—it's speed, customization, and removing the subscription rent that keeps retail traders using inferior tools.