Three AI trading bots named Bizzy, Breezy and Boozy were each given about $333 in USDC and left to trade stocks and crypto around the clock on a Hostinger VPS. The creator built the whole system in a day using Claude Code, Anthropic terminal-based coding agent, and a low-cost AI decision engine.
What Is the 3-Bot AI Trading Bots Experiment?
The experiment gives three AI trading bots a shared $1,000 stake and lets them trade stocks and crypto autonomously for 30 days, after which the creator has pledged to publish real numbers. Each bot, nicknamed Bizzy Bee, Breezy Bee and Boozy Bee, holds roughly $333 in USDC, a USD-backed stablecoin issued by Circle, inside its own exchange sub-account. One bot is framed as the loser that will be shut down at the end.
All decisions come from a single external AI decision engine the creator calls Jev, which he credits with about 300 decisions per minute at a cost he estimates between $3 and $5 per day. Those figures are creator-reported claims from his own setup, not independently verified measurements. The bots cover 537 selectable coins and stocks, chosen by the creator, so the agents can pick anything from Bitcoin to small meme tokens.
Two caveats matter before copying anything here. The video is explicitly labelled entertainment and education, not financial advice, and by the creator's own account the bots were still in dry-run mode when the video was published on 2026-09-25. No profit figures existed at publication; the first real results are promised roughly 30 days later.
How the Trading Loop Works Every 60 Seconds
The bots trade through a loop that repeats every 60 seconds. A Node.js TypeScript service pulls market metrics from the exchange's API, packs them into a small token snapshot that includes the bot's strategy text and current positions, and sends it to the decision engine. The engine returns one choice from a fixed menu of actions, with probabilities, which passes through a risk layer before any market order is placed.
Everything is logged to a database and streamed as events to a read-only React dashboard showing profit and loss, fees paid, engine costs and the live decision feed. Ranking candidates by momentum, attention, news and a volume-score fallback happens in the service code; the model only picks from the menu it is given. This separation is the reason one engine can drive three bots at high frequency for pennies per decision.
The creator states the engine accepted about 100 test calls against live snapshots during the build with all calls succeeding within timeout. That is a first-hand report from his own machine, and it describes API reliability, not trading performance.
Which Safety Layers Stop the Bots Losing the Money?
Four separate controls limit what the bots, and any attacker, can do with the funds. The most important is that API keys can trade but cannot withdraw, so even a successful prompt injection that told a bot to empty the wallet would fail. The creator names the realistic worst case himself: a stolen key could place bad trades, but could not move money out.
The Hostinger VPS deployment adds one more layer: the creator scoped his live exchange keys to the VPS IP address, so only that server can use them. The code also ships with paper trading on by default, meaning no real money moves unless the operator explicitly flips the switch.
| Control | Mechanism | Limits what? |
|---|---|---|
| No-withdrawal API keys | Exchange permissions allow trade, block withdrawal | Theft of funds via prompt injection |
| Sub-accounts | One owner account, three separate balances | Cross-bot losses and mixing |
| IP-bound keys | Live keys restricted to the VPS IP | Stolen keys used elsewhere |
| Paper trading default | Dry-run mode unless manually switched on | Accidental real-money trades |
Note what these controls do not cover. They do not protect against bad trades, exchange insolvency, or the loss of money already on the exchange, and the creator transfers his $1,000 onto the exchange itself. The controls limit theft and runaway automation; they are not a guarantee of trading safety.
How Claude Code Built the Whole System
Claude Code, Anthropic agentic coding tool that runs in the terminal, wrote the entire stack from a short prompt plus an AGENTS.md brief. The build prompt asked for a Node TypeScript service with three profiles, the 60-second metric loop, a fixed-choice decision menu with probabilities, a risk layer, market orders only, database logging and an event stream for a dashboard.
The creator reports the build took around 20 minutes, after which he asked for a read-only dashboard, which appeared after about 8 more minutes with Claude Code screenshotting its own output to check the layout. It also packaged the services as Docker containers locally before any deployment, so the same containers could move to production unchanged.
Two practical details from the session are worth copying. Secrets, including the decision-engine API key, were kept out of printed output by the brief, and the builder first validated everything in test mode against the exchange's European region with no key, then ran a demo-money round trip for all three bots before any live trading was enabled.
Deploying the Bots to a VPS That Never Sleeps
A laptop cannot host trading bots that trade overnight, so the creator moved the stack to a Hostinger KVM 2 VPS running Ubuntu 26.04, the long-term-support release published in April 2026. He registered the domain BeeBots.tech through the same provider, added an SSH key during setup, and gave Claude Code a Hostinger API token so the agent could package, deploy and publish the site end to end.
The deployment prompt was one line: package the project as a Docker Compose for Ubuntu 26.04 and deploy to the VPS. Claude Code first verified the full stack locally in Docker Desktop, then deployed after a single go-ahead. The dashboard went live in dry-run mode first, with live trading enabled by an explicit follow-up instruction.
Ongoing costs stated in the video are about $100 per month for the decision-engine credits, the VPS fee, and a free domain for the first year. These are the creator's own figures for his configuration and plan, and hosting costs will differ by region and provider tier. The repository he shares includes a deploy-to-Hostinger button with paper trading enabled by default.
Frequently Asked Questions
- Are these AI trading bots profitable? No result was published when the experiment started on 2026-09-25. The creator says he will return with real numbers about 30 days later. Until then, any profit claim about these specific bots is unsupported.
- Can the bots withdraw the money? No. Each bot's exchange API key has trading permissions but withdrawal is disabled, and the creator reports the live keys are additionally restricted to the VPS IP address. A compromised key could place bad trades but could not move funds off the exchange.
- How much does the setup cost to run? The creator estimates roughly $3 to $5 per day for the decision-engine calls, about $100 per month in engine credits at his planned volume, plus the VPS and a domain free for the first year. All are his reported figures for his own configuration, not fixed prices.
- Do the bots trade real money by default? No. The creator states the code runs in paper-trading mode by default and no real money moves until the operator explicitly enables live mode. His own live deployment was switched on deliberately after a demo-money test passed.
- Can I build the same thing myself? The creator publishes the code and prompts as free and open source through his community links, including a one-click deploy option for Hostinger with paper trading on. You would still need your own exchange account, API keys and decision-engine credits.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits