Research to production

For quant researchers and academics who run many RL experiments and want a clear path from backtest to live.

Reproducible experiments, then promote

Run experiments with versioned config and full run tracking. Compare runs, pick your best model, and promote it to a live deployment—no separate production pipeline to build.

Every experiment and run is versioned. Config, metrics, and artifacts are stored so you can reproduce results and audit what was trained. When you’re ready, promote a specific model (or the current best) to a live deployment; timeframe and universe come from the experiment, so research and production stay aligned.

Read the research-to-production guide

One broker for research and live

Training and live trading use the same broker account. Data and execution stay consistent; no mismatch between backtest data and live execution.

Configure your experiment with a broker account for OHLCV data. When you promote to live, the same broker is used for execution. You avoid the usual split between “research data” and “live data” and keep one source of truth from experiment to production.

No infra to rebuild

The platform runs your agent in the cloud on a schedule. You don’t need to stand up servers or rewrite pipelines to go live.

Create a deployment, link your broker account, and the platform runs inference and places market orders on schedule. Focus on research and model selection; deployment and execution are handled for you.

© Kabu. All rights reserved.

Built with 🩵 by Zima Blue