OFFLeash: Guardrails for AI Trading Agents

OFFLeash is an autonomous trading agent built to answer a critical question: what happens when AI agents inherit the behavioral mistakes of human traders? OFFLeash autonomously trades a momentum strategy through Alpaca Paper Trading, generating signals every 5–15 minutes while the market is open. Before any proposed trade is executed, it passes through four deterministic behavioral guardrails: oversized positions, overexposure, revenge trading, and overtrading. The key innovation is that the guardrails are completely independent of the LLM. Decisions are made using fixed, auditable rules—not language-model judgment. A human can override a blocked trade from the dashboard, but the autonomous agent cannot override its own guardrail. OFFLeash also measures the counterfactual impact of restraint by tracking what blocked trades would have earned or lost, turning “doing nothing” into a measurable P&L outcome. Its behavior gap compares actual trading performance against a passive buy-and-hold baseline to quantify the cost of selling decisions. The platform includes a real-time dashboard, trading journal, analytics, decision history, natural-language chat, configurable automation, CLI tools, and an MCP server. All trading is paper-based, making OFFLeash a practical environment for testing whether autonomous agents can trade while remaining behaviorally accountable.