Positioning Jev Trader, by jarrodwatts, is an experimental AI trading bot on Monad: it asks a TypeSafe Jev model for one buy-or-sell decision per block while watching the Kuru MON-USDC order book.
What it does Each loop reads the order book in one eth_call, sends the decision question to Jev through the AI SDK's experimental_evaluate on a JevModel, and posts the result as a post-only limit order on that side, one tick inside the touch, in the same batchUpdate that cancels the previous resting order. The model is asked about the move over HORIZON_BLOCKS (default 100, roughly 30 seconds) and returns buy or sell with probabilities; a hold appears only when the block was missed. Written in TypeScript and run with Bun, the bot streams every block over SSE and serves a dashboard with snapshots, the last 1,000 block events, and fill events.
Characteristics The hot path is budgeted at about 300 ms per Monad block and makes exactly two RPC round trips — one book read and one send — using static type-2 fees and a local nonce, with the book read routed through a second RPC endpoint. Without a PRIVATE_KEY it dry-runs: real book, real decisions, simulated fills. MODEL=jev with TYPESAFE_AI_API_KEY selects Jev; the default mock is a momentum heuristic stand-in. The project is MIT-licensed, created 2026-09-16, and its README states that it does not demonstrate a profitable strategy.
When to use Suited to studying how one typed decision can sit inside an automated execution loop — order book in, calibrated answer out, order on the book — with a dashboard and a dry-run mode for observation. It is not presented as a profitable strategy, and the default deployed instance runs the mock model. Any live deployment carries both execution risk and model risk.