Jarrod Watts Ships Jev Trader to Make AI-Guided Crypto Trades in 100ms
An open-source TypeScript bot pairs TypeSafe's Jev decision model with Monad's 300ms blocks to fire a real buy or sell every tick.
Takeaways
- Jarrod Watts open-sourced jev-trader, a TypeScript bot doing one AI decision per Monad block
- Uses TypeSafe AI's Jev System One model for typed buy/sell probabilities in under 100 ms
- Hot loop is two RPC calls: one book read, one raw transaction send, no gas estimate
- Reported latency: p50 18 ms for the order book read, p50 100 ms whole loop
- Dry-run mode works without a private key; live mode needs a TypeSafe API key
- MIT licensed, Bun-based, with SSE dashboard streaming every block and fill event
Developer Jarrod Watts has released jev-trader, an MIT-licensed TypeScript reference for connecting a low-latency decision model to Kuru’s MON-USDC order book on Monad. The Bun application reads market state, requests a buy, sell, or hold decision, and can submit an immediate-or-cancel order within Monad’s roughly 300 millisecond block cadence.
The application processes every block, although fresh model inference runs every 10 blocks by default. Each forecast covers the next 100 blocks, or about 30 seconds. Intermediate blocks retain the current decision without submitting additional orders.
Typed output keeps parsing out
TypeSafe AI’s Jev model accepts structured state and returns typed choices with probabilities instead of free-form text. The company describes it as a System One model trained through reinforcement learning for calibrated decisions, meaning its reported confidence is intended to track observed outcomes. TypeSafe provides more detail in its Jev technical overview.
Jev Trader sends the current order-book state and asks whether the midpoint, halfway between the best bid and ask, will rise or fall over the configured horizon. The response contains probabilities for buy, sell, and hold, along with measured inference latency. The implementation also exposes an upIn10 value equal to the buy probability; developers changing the forecast horizon should account for that field name.