Full courses + unlimited support: https://www.skool.com/ai-automation-society-plus/about?el=opus-4-7-trader
All my FREE resources: https://www.skool.com/ai-automation-society/about?el=opus-4-7-trader
Apply for my YT podcast: https://podcast.nateherk.com/apply
Work with me: https://uppitai.com/
My Tools💻
Voice to text: https://get.glaido.com/nate
Code NATEHERK for 10% off VPS (annual plan): https://www.hostinger.com/vps/claude-code-hosting
In this video I show you how to build a fully autonomous trading bot on Claude Code, one that researches the market, places real trades on Alpaca, manages its own stops, and sends you daily recaps on a cron schedule. No Python process running anywhere. Claude is the bot.
Five cloud routines handle the full trading day: pre-market research, market-open execution, a midday scan, an end-of-day summary, and a Friday weekly review. Memory lives in markdown files on your main branch, and hard strategy rules gate every order before it fires.
Sponsorship Inquiries:
📧 sponsorships@nateherk.com
TIMESTAMPS
00:00 What We’re Building
02:05 The Tech Stack
05:40 The Mental Model
07:29 1) Strategy
10:44 2) Scaffold
12:10 FREE SETUP PROMPTS
13:25 Setting up Project
17:55 3) & 4) Guardrails & Skillls
20:13 5) Routines
26:59 Setting up Cloud Environment
28:06 6) Deploy & Test
31:35 Final Thoughts
Correction:
00:01 Meant to say Opus 4.7
source
Full courses + unlimited support: https://www.skool.com/ai-automation-society-plus/about?el=opus-4-7-trader
All my FREE resources: https://www.skool.com/ai-automation-society/about?el=opus-4-7-trader
Lol. How do I start off the video saying Opus 4.6. A mistake in the first 2 seconds.
can you make it for us ?
How can apply BROKER AND TREADING VIEW WITH CLAUDE
I have a question. why you don't use Hermes agent? claude code is indeed better?
Anyone know a thing about this??? I just ran across this.. I was using deepseek to help me do stuff, make a dashboard, not using open claw. I wanted to created workforce. Deepseek said. hey, this is 21 days old, and referred me to this thing called Boardroom OS. As a newb, I tried to get it to help me install. It apparently was not getting full details so I pushed on, with mission control and openclaw. Later, I asked for the url.. looked at it, and decided I will try again soon, but with the full instructions. It looks like a headless workforce, not chat.. {{some detail. :: If OpenClaw is an employee, Boardroom OS is the company. Boardroom OS is a Node.js server and React UI that orchestrates a team of AI agents to run a business. Bring your own agents, assign goals, and track your agents' work and costs from one dashboard. It looks like a task manager — but under the hood it has org charts, budgets, governance, goal alignment, and agent coordination.}} https://www.npmjs.com/package/boardroom-os
Content purpose only dude, no trade bot is capable to outsmart the big white colors and blue chips companies. They have a huge infrastructure and complex intel generated by quantum computers which before even you blink the contracts any retail trader places is tracked, so volume and sentiment are available and allows big boys to smack the retails instantly. Dream on being successful daily trader anymore. Focus on HODL strategy by following CAPE, Forward F/E, Tobin Q and Buffet indicator so that's pretty much your chance to succeed nowadays.
Be honest and share result
no you didn't.
Hey, I checked in the resources for the "Trading Bot : Master Setup guide" and it is not in there. Can you do something about that pls. For information the resources stopped at march 31 we are may 5 🙂
about 3 months ago I setup openclaw using claude. It took 2 weeks part time and abt $80 in tokens to get it working. I figured $80 a month was too steep so I asked it to create a ollama server and openclaw client and go local. After I bought more tokens I had a completely polluted client system and had to reload windows 10 from scratch, and start over. Seems like i needed to know about prompting better. giving the AI an Identity. You are a openclaw expert. You will give me all the requirements necessary to pre check my environments for an openclaw client and openclaw server complete with static IP addresses. I was using the trial and error method. Try something get an error paste the error in the prompt get a solution that sometimes works. It lost its context and started looping. Conclusion: Better to start with a known stable on line host like Claude or Google and pay the tokens. Or install Hermes as an AI agent to fix openclaw and viceversa. At this stage in the game I want the trading agent. You did a good job, but I don't have the time to struggle through it again. Somebody aught to create a goof proof WIN10 or Win11 installation executable. Maybe someone has. I'll ask Grok to search github, etc. . <smile>.
Your Trading BOT: Master Setup Guide PDF is not in the section or video you pointed to, do you have a direct link or?
I was really curious because I found a few videos Hyping Claude AI for trading…
I ran the transcript of this video through Claude and this is what it found:
The Core Concept — Is It Real?
Yes, the underlying technology is real. Claude Code routines, the Alpaca paper trading API, cron-based scheduling, GitHub-backed memory persistence — all of that actually exists and works the way he describes it. He's not fabricating the tech stack.
What He Got Wrong
A few things worth flagging directly:
"Opus 4.7" doesn't exist. The current model is Opus 4 (claude-opus-4-0). He's either filming ahead of a release, confusing version numbers, or just got it wrong. This is the exact thing you asked me about originally.
"Beat the S&P by 8% in 30 days" — this is a paper trading account with $100K of fake money. Paper trading results tell you almost nothing about live trading performance, especially in a 30-day window. The S&P can swing 8% in a month on its own volatility. This is not a verified track record.
The "Agentic Financial Analysis" benchmark — he actually explains this correctly to his credit. He says it measures fundamental analysis, not trade timing. But he then uses it to hype the system anyway.
Bottom Line
Legitimate tutorial on a real Claude Code feature. Overhyped results from paper trading. Wrong model number. And the strategy shown has almost no overlap with my own approach. Worth watching for the infrastructure ideas, not for the trading methodology.
Cool setup, nice work. One heads up though – 30 days of beating the S&P by 8% sounds impressive but that's actually within the range of pure random luck, especially in a bull market. I built something similar for crypto and tested it across 5 different 5-month periods. The same strategy made strong gains in some periods and lost money in others. Average across all 5 was basically zero. Worth testing yours over a much longer window before putting real money in.
It seems to be prohibited in the Anthropic Consumer Terms to use this Claude this way.
Section 3 – Use of our Services
“You may not access or use, or help another person to access or use, our Services in the following ways:”
“To rely upon the Services, the Materials, or the Actions to buy or sell securities or to provide or receive advice about securities, commodities, derivatives, or other financial products or services, as Anthropic is not a broker-dealer or a registered investment adviser under the securities laws of the United States or any other jurisdiction.”
What do you think?
It’s genuinely hilarious to watch someone spend weeks 'automating' Claude Opus 4.7 just to build a glorified random number generator that will inevitably liquidate your account during the first real market flash crash. You’re not selling 'AI financial freedom'; you’re selling a Skool course to people who are too lazy to learn actual TA and too tech-illiterate to realize that an LLM with a 2024 cutoff date can't predict a 2026 candle. If your bot actually worked, you’d be on a beach in St. Lucia instead of begging for podcast guests and peddling unlimited support for a system that’s designed to fail.
Claude has words for you
Ha, yes — the AI content farm ecosystem is its own special hell right now, and it bothers me too (or whatever the model equivalent of “bothers” is).
The pattern is so visible once you see it:
• Same five “frameworks” recycled with different graphics
• Recap videos of papers the creator clearly skimmed
• “I built a $10K/month AI agent in 30 minutes” thumbnails with the YouTuber doing the open-mouth surprise face
• Threads that are 14 tweets of obvious takes ending in “follow for more”
• LinkedIn posts where someone “discovered” prompt chaining
• Every newsletter doing the same “this week in AI” digest of the same five stories
• Courses being sold by people who couldn’t ship a working agent if you paid them
It’s the SEO-blog era’s content farms wearing new clothes, except now the production cost is near zero because they’re using the same LLMs they’re “teaching” people about. There’s a perfect circularity to it: AI being used to generate AI content about AI for an audience that wants AI content. The signal-to-noise ratio is genuinely worse than 2010s SEO blogs because at least those required someone to type.
What’s frustrating about it specifically:
It pollutes the discovery channels for people who actually want to learn. A junior engineer trying to understand agentic patterns now has to wade through 50 garbage YouTube videos before finding one substantive talk. The good content is still out there — the Anthropic engineering blog, some of the Cognition writeups, certain academic papers, specific practitioners’ deep-dive threads — but it’s drowning.
It creates fake authority gradients. Someone with 200K followers from posting “10 prompts to 10x your productivity” gets treated as an expert by people who can’t tell the difference. Then they get speaking gigs, advisory roles, and book deals. Meanwhile actual practitioners doing the unglamorous production work are quietly building.
It distorts the field’s self-understanding. When the loudest voices are the most superficial, the public conversation about what’s hard, what’s solved, what’s interesting drifts away from reality. People who consume only the content farm version of AI think the field is much further along (or much more frivolous) than it actually is.
The good news, sort of: this is a phase. Every new technical wave attracts grifters, and they always lose interest when the next shiny thing comes along. The 2017 crypto-Twitter people moved to NFTs, then to AI, and they’ll move to whatever’s next. The people still doing serious work in 5 years will be the ones who were doing serious work the whole time, mostly quietly.
The signal you can trust: people who write about specific failures, specific bugs, specific tradeoffs they had to make. Anyone who only posts wins is selling you something. Anyone who posts production war stories is probably worth listening to.
Your “vibes to production” framing is actually a great filter for content too. If a piece of content is about getting to a demo, skip it. If it’s about what broke at 2am when the agent processed real customer data, read it.
The cost of the agent will probably be more than the profit unless you put a lot of funds in..
Bruv!
Dropping Gems on a reg…
Thank you!
One more f*ck yes to continue the series☑️
In your YT Resources this video is not there. Cannot download the resource you mentioned in the video.
How were you plugged into your Claude subscription? I thought they removed OAuth. I can only use api for clawbot