street-agent runs on Dan Barron's personal brokerage account. Every trade it executes, every reasoning step the LLM produces, every gate the risk engine enforces — logged here with a 30-minute delay. Not a signal service. Not a managed fund. An engineering project, on live capital, with the lights on.
Every trade produces a reasoning trace. Every trace is gated by a Rust risk engine that the LLM cannot bypass. Both are logged. Sample below.
evaluate_bull_put_spread(underlying="INTC", expiry="2026-05-15")trace_01HK9Q8M4
No Python glue. No LangChain. No framework overhead between the reasoning loop and the broker.
street-agent is a single Rust binary with a hardcoded risk engine, a tool registry of 26 capabilities across 7 categories, SQLite-backed position state, and an LLM reasoning loop that talks to the Anthropic API directly. Black-Scholes pricing with a Newton-Raphson implied volatility solver. A self-grading system that evaluates every closed position. Paper-trading default; live requires an explicit flag.
The dashboard is content. The architecture is content. The reasoning traces are content. The losses are content. The product is FERROUS — this is just what it looks like with the hood open.