Jev 做 agent 工具路由：返回类型化选择而非文本

Oliver Prompts @oliviscusAI原推发布于 2026.09.20

在 X 上打开原推

you don't need a full llm call just to decide which tool an agent should use next. TypeSafe's Jev returns a typed choice, score, or yes/no instead of generating text, so it can't hallucinate. up to 193x faster and 444x cheaper than a frontier llm on routing tasks. learn more: http://typesafe.ai/blog/introducing-system-one-models-and-jev follow me for more useful tools like this.