Jev 做模型与技能路由：一请求编译出执行方案
输入一句需求，Jev 判断意图与约束、并行扫描 32 个技能，再决定用 GPT-6、Claude 还是 Gemini，并输出匹配分与完整提示词。
I just built an AI router that decides which model and skill should solve your task before you even open ChatGPT. 1 request → 32 skills → 4 AI families → one optimized execution route. You describe the outcome in plain English. Jev analyses the intent, detects the domain and constraints, scans every available skill in parallel, then decides whether ChatGPT, Claude, Gemini or Grok is the strongest model for the job. Ask it to debug a Python API: > skill → Debugging Agent > model → GPT-6 Astra > match score → 91/100 Ask it to build an AI landing page: > skills → Website Builder + UI/UX Designer + Brand System > model → Claude Sonnet 5 > route → design and development stack Ask it to research a market: > skills → Research Analyst + Deep Research + Competitor Analyst > model → best available research model > output → evidence-backed report with primary sources It doesn’t just recommend an AI. It compiles the entire job: role → mission → skill instructions → model context → workflow → quality bar → final output format. Most people choose one AI for everything and wonder why the results are inconsistent. The model was never the only problem. The task was routed to the wrong intelligence with the wrong tools and an incomplete prompt. One request in. The right skill, model and production-ready prompt out. This is what using AI looks like when you stop guessing which model to open. Should I make the Jev Skill Router public? watch the full route compile below ↓
这条是引用着另一条写的：「基于 Jev 架构的 10 步编码 agent 省钱蓝图」