FinRL open‑source trading stack
An open‑source FinRL framework launched, offering an end‑to‑end reinforcement‑learning infrastructure for trading (stocks, crypto, FX) from research through live deployment — a practical toolkit for portfolio projects that mimic institutional pipelines. The release includes training and deployment components useful for hands‑on RL backtesting and execution experiments. (x.com)
FinRL‑X’s system paper, titled “FinRL‑X: An AI‑Native Modular Infrastructure for Quantitative Trading,” was submitted to arXiv on 22 Mar 2026 and listed as accepted at the DMO‑FinTech Workshop (PAKDD 2026). (arxiv.org) The paper formalizes a “weight‑centric” strategy interface that outputs portfolio weight vectors to preserve execution semantics across research, backtesting, and brokered deployment. (arxiv.org) The official implementation lives in the FinRL‑Trading GitHub repository (FinRL‑X) under an Apache‑2.0 license, with the project showing ~256 commits, ~2.8k stars and ~869 forks on GitHub at the time of release. (github.com) The FinRL‑Trading README lists concrete engineering components — Alpaca integration for paper and live orders, multi‑source data connectors (Yahoo/FMP/WRDS), a professional backtest engine built on the bt library, SQLite-based data persistence, and Pydantic settings with Python 3.11+ as a prerequisite. (github.com) The paper and repo both state explicit support for hybrid pipelines that combine rule‑based modules, reinforcement‑learning allocators, and LLM‑derived sentiment signals while keeping downstream execution unchanged. (arxiv.org) (github.com) The repository includes a FinRL_Full_Workflow.ipynb demonstrating a research‑to‑paper‑trade workflow and lists example configurations (API keys, Alpaca account) that make reproducing an end‑to‑end experiment straightforward for portfolio projects. (github.com)
