User repository on GitHub
Three AI trading bees on OKX, every decision by Jev. Paper trading by default. Not financial advice.
- Forks
- 126
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Beebots
Three AI trading bees race each other on OKX perpetual futures. Every decision comes from Jev picking one move from the moves that are legal right now, and every order then passes a risk layer written in plain code. Paper by default, with real market prices.
- Stars
- 269
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- imikerussell
- Last verified
- Oct 2026
About this project
Three AI trading bees race each other on OKX perpetual futures. Every decision comes from Jev (TypeSafe AI's decision model), and every order goes through a risk layer written in plain code. A live dashboard shows each decision, order, fee and funding payment as it happens. It runs on paper by default. The bees use real market prices and simulated money. Nothing touches an exchange account unless you change the settings yourself, on purpose. Use code MAGIC10 at checkout for 10% off.
Quoted from the project's README. Read the full README.
- Languages
- TypeScript
Highlights
- Runs on paper out of the box: real OKX market prices, simulated money ($333 per bee), and no exchange account touched unless you change settings.
- Deploys from a single docker-compose.yml on any machine, or through a Hostinger button that opens a two-hour Setup window.
- Three built-in trading styles — Breakout on BTC/ETH/SOL/HYPE, Trend on BTC/ETH, Momentum on any liquid coin — chosen from the coins your bee may trade.
- The Hive leaderboard is opt-in and shows percent gain only; the engine refuses to report in MODE=live and the board rejects live reports.
Watch out
The README is written as a launch page, and the one-click deploy button is a Hostinger affiliate deep link carrying offer and affiliate parameters. Setup has no authentication of its own: the page is open to whoever reaches the server first, it closes for good once you press Start, and it locks by itself after a two-hour window, so anyone who wins that race can replace your keys. Eight wrong owner-password tries lock you out for fifteen minutes. The Hive shares more than the percent figure suggests — bee names, styles, trade counts, equity, funding and every paper fill, so results can be replayed against public OKX prices and marked verified — while keys, exchange data and server address are stated as never shared. The optional Beekeeper Zap can rewrite a bee's rules and coin list once every twenty hours, with Undo available. Jev spending is capped at two dollars a day by default, and the OpenAI key is needed only to design bees and paint portraits. Nothing in the README states what the risk layer's caps and stops actually are, or how a strategy behaves when a coin fails the volume and spread gate and the bee simply waits. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Paper is the default, and the shared leaderboard refuses live runs: the engine will not report in MODE=live. But the README is explicit that switching to real money is a setting you change yourself, on purpose, and is then your decision and your risk. It ships with no warranty and calls itself an experiment, not a trading product. There is no credential scoping or withdrawal limit described, so if you do flip the mode you are relying on your own reading of a risk layer the README never spells out.
The loop is worth reading even if you do not adopt it. Each tick the engine reads live OKX data (tickers, candles, RSI, MACD, ATR, Bollinger, Donchian, funding, open interest), summarises the market and the bee's own position numerically, then asks Jev to choose one move from the menu of currently valid moves, with probabilities. The bee's written rules go into that call, so they steer the choice but cannot invent moves the style does not offer. No accuracy figure, backtest or non-Jev comparison is published.
You install the repository's docker-compose.yml onto a VPS you choose, and Hostinger takes an affiliate cut, which the badge URL spells out in its own query parameters. The same file runs standalone with curl and docker compose up -d. Worth knowing before you start: the setup page is unauthenticated until you finish it, the owner password only gates later changes from the public dashboard, and the README suggests a domain with HTTPS so your Jev and OpenAI keys do not travel in plain text.
Stops, caps and a never-flat-for-long rule apply to every bee and are documented in strategies/DRAMA_RULES.md, which the README points to rather than inlines, so the numbers themselves are not in the README. Jev spending is capped at two dollars a day by default, and the Beekeeper may never touch leverage, stops, caps or real-money settings. Each bee starts with 333 dollars of paper money. No drawdown statistics or historical results are published, so there is no basis for expecting profitable behaviour.
Three built-ins. Breakout, the original bee Bizzy, takes one volatility breakout a day on BTC, ETH, SOL or HYPE and rides it to the daily close. Trend, Breezy, follows trend on BTC and ETH only, with few trades and volatility-based sizing. Momentum, Boozy, chases the strongest 7-day mover across every liquid coin and adds to winners. Your sentence picks which coins a bee may trade, and that limits which style it can run on.
A bee's rules text and its coin list, once per bee every 20 hours, and never leverage, stops, caps or real-money settings. It runs as a Zap on Zapier where Jev picks which of the three bees to look at and Claude Opus 5.5 writes the replacement rules. Every rewrite appears on the dashboard with an Undo next to it, and Undo, Connect and Disconnect all require your owner password. It works on paper, so no exchange account is involved.
New versions ship as GitHub releases and your dashboard shows an Update available link next to the trading mode rather than updating itself. To update you pull the new images and bring them up, and your bees, settings and history live in Docker volumes and are kept. Set UPDATE_CHECK=false to turn the check off. If your bee's named coin fails the volume or spread gate, which means at least a million dollars of 24-hour volume and a tight spread, the bee simply waits until it passes.
More Jev resources like this
123GitHub stars
Electron desktop browser where Jev picks the next executable action from a page's observed controls while a separate text model plans; deterministic code builds the action space, validates responses, dispatches input and keeps receipts.
Jev Demos & Experiments#community#jevbest
Community
michaelswissa/jevryTypeScript
michaelswissa/jevryTypeScript75GitHub stars
A Node package that distils Jev-style typed answers into small local classifiers, trained either from saved labels or from a Jev teacher, then served in Node or a browser worker.
Jev Demos & Experiments#community#jevbest
Community
AndrewPrifer/jimothyTypeScript
AndrewPrifer/jimothyTypeScript64GitHub stars
Turns a Playwright trace into a narrated demo video through a chainable pipeline - speed control, subtitles, TTS, click and cursor overlays, zoom - with ffmpeg doing the render.
Jev Demos & Experiments#community#jevbest
98GitHub stars
Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
1Panel-dev/laya-serverPython
1Panel-dev/laya-serverPython38GitHub stars
Standalone CLI skill that drives isolated Playwright Chromium with Jev choosing scoped in-page actions, defaulting to Vercel AI Gateway with TypeSafe as an option.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
hqman/jev-browser-skillTypeScript
hqman/jev-browser-skillTypeScript41GitHub stars
Browser-focused TypeScript library for typed decisions that runs open model reproductions on-device via Transformers.js, returning one calibrated distribution per question.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
nico-martin/open-jevTypeScript
nico-martin/open-jevTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.