Jev Trade

September 19, 2026 · View on GitHub

I built a trading bot with Jev. Jev reads the Hyperliquid book every tick and answers buy, sell, or hold. The bot sends the order. Five coins, five wallets, real fills.

Watch the live desk

Jev makes the call. Hold is one of its answers, so a tick can end with no order. Position, balance, and PnL come from Hyperliquid.

Based on jev-trader by Jarrod Watts (MIT). Venue is Hyperliquid, not Monad / Kuru.

What you are looking at

Five isolated sleeves: BTC, ETH, SOL, DOGE, BNB. Each has its own wallet and its own Jev call.
The left pane is the live book as candles. Green and red marks are fills. The entry line is the open position.
The right pane is the latest Jev call and the tape of every tick.
The table is Positions and Trades, the same split a futures desk uses.

A live key on testnet or mainnet sends real orders. Start with a dry run.

How a tick works

The bot reads the book.
Jev picks long or short, then open, close, or hold.
An entry is a post-only Alo quote one tick inside the touch, so it sits on the maker side until a taker hits it.
An exit is an Ioc that crosses the touch and fills on the spot.
Hold sends no order and pulls any resting quote Jev no longer wants.

The bot is Bun on port 3000. The dashboard is Next in web/ on port 3001. Keys, evaluate, and orders stay on the Bun process.

Dry run

Bun 1.2 or newer. No PRIVATE_KEY means a dry run: real book, real decisions, simulated fills. Default MODEL=mock is a momentum stand-in and needs no API key.

cp .env.example .env
bun install
bun run start

Dashboard (second terminal):

cp web/.env.example web/.env.local
bun run dev:web

Open http://localhost:3001. The page reads $NEXT_PUBLIC_API_URL/events (default http://localhost:3000).

Live Jev

Set MODEL=jev and pick an API. Official TypeSafe is the default.

MODEL=jev
JEV_PROVIDER=typesafe
TYPESAFE_API_KEY=
# JEV_MODEL_ID defaults to jev-latest

Get a key from docs.typesafe.ai.

Vercel AI Gateway is still supported:

MODEL=jev
JEV_PROVIDER=gateway
AI_GATEWAY_API_KEY=
# JEV_MODEL_ID defaults to typesafe-ai/jev

If JEV_PROVIDER is unset, the bot uses TypeSafe when TYPESAFE_API_KEY is set, otherwise Gateway when AI_GATEWAY_API_KEY is set.

Live testnet orders

Keep HL_TESTNET=true.
Set PRIVATE_KEY for the first coin (BTC).
Copy .wallets.example.json to .wallets.json and put a key on each other sleeve. Or set WALLETS_JSON.
Get mock USDC from https://app.hyperliquid-testnet.xyz/drip. The faucet only pays addresses that have deposited on mainnet.
Leave DRY_RUN=false. A missing key on a sleeve still dry-runs that sleeve.

HL_TESTNET=false is mainnet. Do not flip that until you mean it.

Tests

bun test

CI runs the same command on push and pull request.

Env

See .env.example. The ones that change behavior:

Variable	Default	Meaning
`HL_COINS`	`BTC,ETH,SOL,DOGE,BNB`	Sleeves to run
`HL_TESTNET`	`true`	`false` is mainnet
`MODEL`	`mock`	`jev` needs a TypeSafe or Gateway key
`JEV_PROVIDER`	`typesafe`	`typesafe` or `gateway`
`TYPESAFE_API_KEY`	empty	Official TypeSafe key
`AI_GATEWAY_API_KEY`	empty	Vercel AI Gateway key
`PRIVATE_KEY`	empty	First coin. Empty is a dry run
`DRY_RUN`	`false`	`true` simulates every sleeve
`TICK_MS`	`2000`	Decision + requote cadence
`PRICE_MS`	`200`	Chart and mid prints. Does not call Jev
`QUOTE_USD`	`40`	Quote notional per tick
`CLOSE_SLIPPAGE_BPS`	`5`	How far an exit crosses the touch
`PORT`	`3000`	Bot SSE

Endpoints

GET / snapshot: model, sleeves, latestByCoin
GET /history last 1000 ticks per coin
GET /tape all-time mid series plus fill marks per coin
GET /events SSE: snapshot on connect (historyByCoin, tapeByCoin), then block, quote, fill keyed by coin

Layout

src/           Bun bot
test/          bun tests
web/           Next dashboard
assets/        README shots of the live desk