Agents waste reasoning on one recurring question: which of their capabilities should handle this next? JevRouter gathers the models, subagents, skills, MCP tools and command-line tools into a single list and asks Jev to pick one.
Jev supplies the choice and its probabilities. JevRouter then applies the rules Jev should not own: what is available, what is permitted, how risky the step is, and when to ask you first.
Comes with a website, a benchmark, a cookbook of examples and a Chinese introduction.
What you get
- Full TypeScript source code, MIT licensed
- Command-line tool, skills and examples
- Cookbook, docs and tests






