tylerbroqs/sturna

Sturna is an open source engine that simulates an investment desk run entirely by six AI agents. Each agent owns one step of the process: Scout finds eligible assets from a tokenized universe of private-market and real-world assets, Analyst scores each one on growth, quality, momentum, liquidity and thesis fit, Architect builds a weighted basket while respecting risk caps, Sentinel independently checks the plan against risk limits and can halt the whole cycle if something looks too risky, Executor turns approved decisions into trade orders, and Steward marks the resulting portfolio to market and watches for drift afterward. By default the project runs in paper mode, meaning every trade is simulated with no real money at risk. It is built to work with Robinhood's official agentic trading MCP, a Model Context Protocol interface, so it can be switched to place real orders once you provide the right credentials, though the README is explicit that this is a research and education project and not investment advice. The code is written in TypeScript with no framework dependencies in its core, so it can run in Node or directly in a browser. You install it with npm, then either start an HTTP API server, run a single cycle from the terminal, or run the unit tests. It also includes a backtesting command that replays a simulated year of weekly rebalances and compares results across different investment mandates, with reproducible results when you set the same random seed. Developers can use every agent as an importable function in their own code, and the repository includes example projects showing a custom agent pipeline and the whole swarm running inside a plain browser page. The project is licensed under MIT.