Autotrader – paper trading AI agent for Indian equities (github.com)

A new AI-driven paper trading agent named Autotrader has been launched for experimenting with Indian equities, specifically tracking the Nifty 500 index. This agent utilizes Claude, running on a free Google Cloud Platform virtual machine, to autonomously execute trades based on real-time price feeds from the Kite Connect API while considering actual transaction costs like brokerage and taxes. The experiment, which began and paused after a two-week run ending May 8, 2026, achieved an 8.05% return on an initial capital of Rs 1,00,000, showcasing its potential in automated trading strategies. The significance of this project lies in its open-source nature, encouraging the AI/ML community to build upon this foundational work. It offers a detailed setup, including essential scripts for price polling, strategy execution, and trading signals, while allowing users to modify only the core trading strategy file. This design not only emphasizes safety by applying real-world trading costs and limitations, such as no short selling and a loss floor, but also facilitates further exploration into the optimization of trading algorithms through shared learnings and logs from the experiment. The project serves as a practical case study on AI's capabilities in the financial domain, particularly within emerging markets like India.