I built an AI crypto trading bot that runs 24/7 on my Mac Mini — here's what I learned.
Built an AI crypto trading bot that runs 24/7 on a Mac Mini. Triple LLM consensus — 3 models must agree before any trade fires. 15 safety checks. Survived every major crash in backtesting. Launching as PassiveBot AI on May 12. $197 once.
Trending on Indie Hackers
The "triple LLM consensus" framing sounds rigorous but LLMs aren't analyzing market microstructure - they're pattern-matching on text. That's a meaningful distinction.
None of this means the project isn't interesting as a technical experiment. But "runs 24/7 on my Mac Mini" and "survived backtesting" aren't the evidence that should make someone hand over $197 for financial automation software. What does live trading performance look like over the last 90 days?
Fair pushback and you're right to make that distinction. The LLMs aren't doing microstructure analysis — they're acting as a consensus filter to reduce false positives from the technical signals. Three models disagreeing = skip the trade. It's a noise reduction layer, not alpha generation.
The more interesting part is the XGBoost meta-learner sitting underneath it. Every trade outcome gets fed back into an ML model that reweights which LLM combinations work best under which market conditions — regime, F&G range, RSI level, time of day. The system literally gets smarter with every trade it takes. That's the part that makes it meaningfully different from a static rules-based bot.
On live performance — we're currently in paper trading phase by design. F&G has been below 15 for weeks and the system is correctly sitting out. No shorts in extreme fear is a hard rule because relief bounces destroy short positions. Zero paper trades in this period is the system working, not failing.
The honest answer on 90 day live performance is: we don't have it yet because we haven't met our own promotion criteria — 10+ paper trades at 55%+ win rate before going live. That's the standard we set before touching real money.
$197 is priced as an educational product — full source code, 14 modules, you own it forever. It's not a managed fund or signal service. The value is in learning to build and run your own system, not in outsourcing your trading decisions to a black box.
Do you have any kind of trial period or format where a potential customer can test the performance of your solution? Because $197 at the start, in my opinion, is a pretty high entry point for an unfamiliar product.
This is a subjective opinion, of course, I may be wrong
The $197 is a one-time payment for the full source code — 14 modules, every script, every safety system. You own it forever. There's no ongoing cost, no black box, no trusting someone else's server.
The value isn't "buy a bot that trades for you" — it's learn exactly how a professional automated system is built, understand every decision it makes, and run it yourself. That's a fundamentally different product to a subscription signal service.
On validation — the system has been backtested across every major crypto crash since 2018. LUNA collapse, FTX, COVID crash, 2022 bear market. It survived all five scenarios. That's not a guarantee of future performance but it's a meaningful bar that most bots never clear.
And there's a 7-day money back guarantee — if you buy it, go through the modules, and feel it wasn't worth it, you get a full refund.