用 Jev 做毫秒级交易决策的三个仓库

ggwp @ggwplabs原推发布于 2026.10.08

在 X 上打开原推

Honestly, this should be f*cking illegal. Your LLM spends 30 seconds explaining a trade. The candle doesn’t wait for the explanation. Jev is built for the decision instead: BUY / SELL / HOLD, with probabilities rather than a paragraph. The original post reports response times of 70–500 milliseconds. Three repos worth looking at: → jev-trader A trading loop on Kuru’s MON-USDC market, making a decision roughly every Monad block. It places and replaces post-only limit orders, aiming to capture the spread. Important detail: the default setup uses a mock model and simulated fills. You need to explicitly configure Jev. → QuantDinger An open-source AI trading platform with Jev integration. Research, Python strategies, backtesting, and paper/live trading across crypto, stocks, and forex. → AIStock A stock-analysis project built around the idea that one person can become their own research desk. The quoted Jev pricing is $0.042 per million input tokens, with free output. But cheap decisions aren’t automatically good decisions. A bad entry in 200 milliseconds is still a bad entry. And model latency isn’t the same as getting an order filled. The setup worth testing: Let the big model research the market. Use Jev for the narrow decision. Keep position limits and execution rules in code. Paper trade it before putting money behind it. Your trading bot doesn’t need to write a beautiful essay while the opportunity disappears. It needs to make a decision you can measure. Save these repos before building another chatbot with a buy button.

这条是引用着另一条写的：x.com/ggwplabs/status/2106004888312217829