A quant factory that wants to live on your server, not theirs

What it does

QuantMind is a self-hosted quantitative research and trading platform that stuffs the entire workflow—market data, automated factor discovery, model training, backtesting, and live execution—into a Dockerized stack. It is built around Chinese retail broker infrastructure like QMT and Tongdaxin, while claiming support for Hong Kong, US equities, futures, and crypto markets. The selling point is privacy-by-default: your data, models, and signals stay on your own hardware rather than a vendor’s cloud.

The interesting bit

The project treats “local-first” as a feature, not a bug. It ships an offline bundle containing pre-built images, historical data, and even pre-trained models so the system can bootstrap without internet access. When you do need fresh ticks, it pulls through a QuantDB API or a Baidu Netdisk archive rather than sending you on an open-source data scavenger hunt.

Key highlights

• Bundles Microsoft Qlib for event-driven backtesting, RD-Agent for LLM-guided factor evolution, and Optuna hyperparameter search behind a FastAPI backend and an Electron frontend.
• Advertises 300+ pre-computed features (L1/L2 microstructure and capital flow) stored in Parquet/DuckDB, plus a model factory covering 13 algorithms from LightGBM to Transformer variants.
• Bridges to live trading via an encrypted WebSocket proxy to the Windows-only QMT (xtquant) retail terminal, complete with order anti-starvation queues and a disconnect watchdog.
• Pushes model-generated watchlists directly into the Tongdaxin desktop terminal as custom sectors and supports “lightning orders” from its intraday radar pop-ups.
• Runs a 7×24 RSS sentiment pipeline that tags entities and quantifies news as bullish or bearish, feeding a market-analysis dashboard alongside the usual P&L screens.

Caveats

• Several execution paths—notably the QMT live-trading agent and Tongdaxin integration—are tightly coupled to Windows and Chinese retail broker software, so traders outside that ecosystem may hit friction.
• The README is long on feature claims but short on reproducible evidence; statements like “millions of quotes load in seconds” and “blockchain support” arrive without benchmarks or architectural detail.
• Offline deployment depends on downloading multi-gigabyte bundles from a project-specific CDN; if that endpoint drifts, the “one-click” experience becomes a manual data-archaeology exercise.

Verdict

Worth a look if you want an on-premise quant workstation tightly integrated with Chinese retail brokers. Give it a pass if you need a broker-agnostic setup, a fully open data pipeline, or a non-Windows execution layer.