Build an AI Crypto Trading Bot That Trades While You Sleep
Build an AI crypto trading bot that only trades overnight: install one skill, backtest, fix a losing strategy, and paper trade it live.
Crypto tends to make its biggest moves at night: the spikes, dips, and bounce-backs that you only discover in the morning when you check the charts and see what ETH or SOL did without you. This guide shows how to build an AI crypto trading bot that trades only during those hours, from 10 p.m. to 6 a.m., so you stop feeling the FOMO every time you go to bed.
The whole thing is built step by step with AI, using the Strategy Creator skill from AlphaInsider. The AI designs the strategy, backtests it, hosts it, schedules it to run every night, and hands it back ready to paper trade or automate with a real broker.
The honest version of this story matters, too. The first backtest lost more money than simply holding the coins. You'll see exactly how one revision prompt cut the drawdown roughly in half and beat every benchmark.
Table of Contents
- Key Takeaways
- Why Crypto Moves While You Sleep
- The Plan: A Bot That Only Trades Overnight
- Why Use the Strategy Creator Skill
- Prerequisites / What You'll Need
- Step-by-Step Guide
- Download the Strategy Creator Skill
- Create a Project and Install the Skill
- Answer the Strategy Questions
- Run the First Backtest
- Revise the Strategy to Break the Correlation
- Compare Version One and Version Two
- Implement the Strategy on AlphaInsider
- Confirm the Nightly Schedule Is Live
- Automate It When You're Confident
- Why Paper Trade Before Going Live
- Tips & Best Practices
- Common Mistakes to Avoid
- Frequently Asked Questions
- Conclusion
Key Takeaways
- Crypto often moves most overnight. This bot is built to trade only from 10 p.m. to 6 a.m., the window when you'd otherwise miss the action.
- One AI skill does the heavy lifting. The Strategy Creator skill builds the strategy, backtests it, hosts it, and schedules it to run automatically.
- The first version failed the backtest. Over roughly one year (370 nights), it lost about 54%, which was worse than the 45% loss from simply holding the five coins equally.
- One revision prompt changed everything. Adding a 50-period moving average filter (longs only above it, shorts only below it) cut the loss to 22.5% and the max drawdown to 31.2%, beating buy-and-hold and Bitcoin.
- Version two sits in cash on 64% of nights. It trades selectively instead of every night.
- Paper trading beats backtesting for proof. A backtest is hypothetical, so the strategy trades a $10,000 paper balance live first.
- Automation is one button away. Crypto brokers start at $100, with up to 10x leverage available if you want it.
Why Crypto Moves While You Sleep
Over the past couple of years, and especially in recent months, a pattern keeps showing up in the crypto markets. After bedtime, volatility picks up: bigger spikes, sharper dips, and faster bounce-backs.
The classic scenario goes like this. You wake up, eventually get over to the charts, and realize ETH already went somewhere big and SOL already made its move, all while you were asleep. You missed the move, and you missed the trade.
Instead of setting alarms or staring at charts at 2 a.m., the fix is to let a bot do it.
The Plan: A Bot That Only Trades Overnight
The goal is a trading strategy with one unusual rule: it only trades while you're sleeping, roughly from 10 p.m. to 6 a.m.
Here's the build setup used in this walkthrough:
- AI platform: Grok, on the $30-per-month plan, using the web platform. Any AI platform works; Grok is just the one used here.
- The skill: AlphaInsider's Strategy Creator skill, installed into Grok.
- The strategy idea: momentum trading on the top cryptocurrencies, overnight only.
- The goal: trade at least a couple of times per week during the night (ideally every night), so live data comes in quickly enough to judge whether the strategy has something to it.
Why Use the Strategy Creator Skill
You could try building this with a plain AI chat. You'd probably spend a lot more time, though. And unless you already know how you'd execute the trades, you won't have an easy, effective way to track the strategy with paper trading or trade it with live funds.
The skill closes that gap. Once it's installed, you call on it, tell it what you want to build, and it asks the right questions to design the strategy for you. From there it will:
- Build the strategy from your answers.
- Host the strategy somewhere so it can run without you.
- Schedule the task on your AI platform so it trades every night automatically.
- Backtest it as well as it can, to give you a sense of how the strategy would have looked.
- Hand it over ready to automate, with up to 10x leverage available. That sounds wild, but it's there if you want it.
Prerequisites / What You'll Need
- A free AlphaInsider account. You'll use it for the skill download, the API key, and tracking the strategy.
- An AI platform subscription that can run scheduled tasks. This walkthrough uses Grok at $30 per month; ChatGPT and Claude work too.
- A strategy idea. It can be as simple as "trade momentum on the top cryptos overnight."
- $100 (optional, later). That's the minimum to start on the supported crypto brokers if you decide to automate with real funds.
- Time: one sitting, mostly spent answering questions and reviewing backtests. A single revision took Grok about four and a half minutes.
- Skill level: beginner-friendly. The AI asks the questions and writes the logic.
Want a bot that trades overnight?
Create a free account, grab the Strategy Creator skill, and build yours tonight.
Step-by-Step Guide
Step 1: Download the Strategy Creator Skill
Start on alphainsider.com and head down to the API Documentation, then open the Agent Skills section. Grab the Strategy Creator zip file and download it.
Step 2: Create a Project and Install the Skill
Before installing anything, create a new project on your AI platform. Do this for every new strategy so everything stays contained in that project folder. If you skip this step, the skill will likely prompt you to do it later anyway, so it's easier to do it now.
Inside the project, upload the zip file and give the AI a prompt like this:
"Install the Strategy Creator skill. I want to build a trading strategy."
The AI installs the skill, confirms the project is ready, and starts asking questions to design a paper strategy. That's always the first stage: once built, the strategy trades paper money so you can track real performance before risking anything.
Step 3: Answer the Strategy Questions
The first round had four questions. Here's how the overnight bot was answered:
- Asset class: crypto.
- Strategy style: momentum on the top cryptocurrencies.
- Decision-making: option C, a hybrid decision-making process.
- Timing: trade nightly, with extra detail about the overnight concept.
This is where you can get as creative as you want. Give the AI as many details, sources, links, resources, and data feeds as you like. Ask whether it can pull a certain kind of data, and how.
The second round covered the finer details:
- Trade cadence: went with the AI's recommendation, which can always be changed later.
- What to trade: the top crypto names the AI suggested.
- Direction: both long and short.
- Weighting: equal weight for now.
- Remaining questions: the recommended options, just to get a feel for the backtest before adjusting.
While the AI works, open the plan.md file in your project's files panel (on the right side in Grok). It shows the strategy's plan as it takes shape. The plan is fully editable: you can come back in two months, edit it, reset the strategy, and end up with something completely different based on the data.
Step 4: Run the First Backtest
When the AI offers a backtest, take it. It gathers whatever data feeds it needs and builds the best backtest it can.
The first results covered roughly the past year, about 370 nights, and they weren't good:
| Metric | Strategy (Version One) | Five-Coin Equal-Weight Buy-and-Hold | Bitcoin |
|---|---|---|---|
| Return | About −54% | −45% | −25% |
| Max drawdown | 61% | 65% | — |
The strategy did worse than simply holding the coins. Looking at the charts made the reason clear: the strategy was highly correlated with the buy-and-hold and with Bitcoin. It was mostly riding the same moves as the market, just with extra risk.
Even so, a failed first backtest still gives you something to work with.
Step 5: Revise the Strategy to Break the Correlation
When the AI asks for next steps, choose the option to revise the strategy and retest. The goal here was to pull the strategy away from the buy-and-hold. The prompt went roughly like this:
"It seems like we're very correlated to the buy-and-hold, and I don't want to be. Can we add a filter where, if price is below the 50 moving average, we only take shorts, and if it's above the 50 moving average, we only go long? And maybe filter out certain nights where we don't trade at all."
This is the core loop of building with AI: prompt, retest, compare, and repeat until you have exactly what you want, or as close as you can get.
Step 6: Compare Version One and Version Two
About 4 minutes and 22 seconds later, Grok came back with the revised version:
| Metric | Version One | Version Two | Five-Coin Buy-and-Hold | Bitcoin |
|---|---|---|---|---|
| Return | −51.1% | −22.5% | About −41% | −23% |
| Max drawdown | 61% | 31.2% | — | — |
| Nights in cash | — | 64% | — | — |
The benchmark and version one numbers shift slightly from the first run because the retest produced fresh figures.
In what was a bear market for crypto, version two technically beats everything. It still lost money, but less than every benchmark. Its max drawdown is about half of the original, and it holds cash on 64% of nights instead of trading every night. On the comparison chart, version two holds up best of all the lines.
The open question is how it performs in a bull market. Crypto appears to be starting to rally, so the next few weeks and months should answer that. The only way to find out is to take it live.
Step 7: Implement the Strategy on AlphaInsider
Pick the next-step option to implement version two on AlphaInsider. The AI will ask for your API key, which it needs to create the strategy on your account. To generate one:
- Make sure you're signed in (a free account is all you need).
- Go to Settings → Developers.
- Name the API key and select AI agent as the key type.
- Enter your password to create the key.
The key is reusable. The one used in this walkthrough was made a couple of days earlier for another strategy. Paste it into the chat, or set it yourself without pasting it.
Then answer a few final setup questions:
- Where it lives: a new cryptocurrency strategy in your portfolio.
- Visibility: public.
- Starting balance: $10,000 in paper money, used only to track how the balance plays out over time.
- Self-healing: on, with notifications only when there's an issue with the strategy.
Saying "recommended all" handles these quickly.
Step 8: Confirm the Nightly Schedule Is Live
The AI creates the strategy, named Overnight Crypto Momentum V2 in this case, and returns the strategy ID plus a direct link. The strategy page shows an explanation of what it does and its schedule.
Back on Grok, open the schedule to confirm the automation: Overnight Momentum V2, daily at 10 p.m., active. It runs live from your AI account, and every time it fires with a trade to take, the trade lands on the strategy.
In your AlphaInsider portfolio, the new strategy appears at the bottom of the list until it starts trading.
Step 9: Automate It When You're Confident
When you're ready for real funds, open the strategy and click Auto Trade. From there you can:
- Add it to an existing trading bot, or
- Create a new trading bot by adding a new broker connection.
The supported crypto brokers let you start with as little as $100. Once the strategy is in a trading bot, trades are taken automatically without you doing anything.
The plan in this walkthrough, though, is to let it paper trade first, gather data, and check performance. Only once there's confidence does it go into a live bot.
Why Paper Trade Before Going Live
A backtest is a hypothetical. It says that if the strategy had traded this way, it could have produced this return. Nothing guarantees it will actually trade that way going forward.
Paper trading with live market conditions is the best way to find out without risking real money. The only trade-off is time: how fast you get useful data depends on how active the strategy is. That's why this bot is built to trade on as many nights as possible (ideally every night, or at least a couple of times a week) so the evidence builds quickly.
Tips & Best Practices
- One project per strategy. It keeps files, plans, and schedules contained, and it avoids a prompt from the skill later.
- Feed the AI everything you've got. Sources, links, data feeds, and specific rules all sharpen the strategy.
- Watch the plan.md file as it builds. It's the clearest view of what your strategy actually does.
- Check correlation, not just returns. A strategy that mirrors buy-and-hold adds risk without adding edge.
- Use "recommended all" when the AI's defaults look reasonable. It keeps the build moving.
- Reuse your API key across strategies instead of generating new ones.
- Treat leverage as optional. Up to 10x is available, but nothing requires you to use it.
Common Mistakes to Avoid
- Taking a failed backtest as the final answer. Version one lost more than buy-and-hold, and one revision turned it into the best performer of the group.
- Taking a good backtest as proof. Version two still needs live paper trading, especially to see how it handles a bull market.
- Skipping the new project step. Loose chats make strategies harder to organize and maintain.
- Going live on day one. Let the paper balance build a real track record before connecting a broker.
- Forgetting the plan is editable. If the data tells you something new, edit plan.md and reset instead of starting over.
Frequently Asked Questions
What hours does the overnight bot trade?
Roughly 10 p.m. to 6 a.m., the window when the video's creator is asleep. The automation is scheduled to fire daily at 10 p.m.
Which AI platform was used to build it?
Grok, on the $30-per-month plan through the web platform. The Strategy Creator skill also works with ChatGPT, Claude, and other AI platforms.
How did the backtest perform?
Version one lost about 54%, worse than the five-coin buy-and-hold's 45% loss. After one revision, version two lost 22.5% with a 31.2% max drawdown, beating buy-and-hold (about −41%) and Bitcoin (−23%) during a bear market.
What changed between version one and version two?
A 50-period moving average filter was added: long trades only above it, short trades only below it. There was also a filter to skip certain nights. As a result, version two holds cash on 64% of nights.
Why paper trade instead of trusting the backtest?
A backtest is hypothetical. Paper trading in live market conditions shows how the strategy actually trades, without risking real money.
How much money do I need to automate it?
The supported crypto brokers let you start with $100. The strategy itself tracks a $10,000 paper balance, which isn't real money.
How much leverage can I use?
Up to 10x is available for automation. It's there if you want it, but it isn't required.
Do I need to know how to code?
No. The skill asks the questions, writes the logic, runs the backtest, and deploys the strategy. You answer in plain language.
Can I change the strategy later?
Yes. The plan.md file is fully editable. You can come back months later, edit it, and reset the strategy based on new data.
Conclusion
Building an AI crypto trading bot that trades while you sleep comes down to a handful of steps. Install the Strategy Creator skill, describe your overnight idea, and backtest it. Then revise, and keep revising, until it actually has an edge. The first version here failed outright, and a single prompt about correlation turned it into the strongest performer in the group.
Now it paper trades every night starting at 10 p.m., building a live track record before real money gets involved. Create your free AlphaInsider account, grab the Strategy Creator skill, and build a bot that trades the hours you'd otherwise miss.
