In this video I show you how I turned Claude Opus 4.7 (correction: I accidentally said 4.6 at the start) into a 24/7 trading agent using Claude Code routines. The whole point is simple: Claude is the bot. No Python process running on a server somewhere. Each routine “wakes up” on a cron schedule, reads the same memory files, does its job (research, decisions, execution), then writes back lessons so the next run stays disciplined instead of acting stateless. I walk through the exact stack: Claude Code routines as the scheduler, Opus 4.7 as the model, Alpaca as the brokerage (paper or live), Perplexity for research, and ClickUp for daily recaps. Then I lay out the mental model that makes this work—memory architecture in markdown files, hard guardrails before any order fires, and treating tokens like money because context budget and context rot are real. Finally, I break the trading day into five cloud routines: pre-market research, market-open execution, a midday scan, end-of-day summary, and a Friday weekly review. I also explain the difference between local vs remote routines (and why remote needs a GitHub repo so changes can be committed back to main), plus the guardrails I use to keep an “eager” autonomous agent from going off the rails.