jevort installs into pi as a small extension and switches reasoning effort automatically on every turn, so a quick factual question doesn't pay for the same effort as a hard multi-step task.
The decision comes from Jev reading the prompt and its context, then picking a model-compatible effort level before the turn runs, rather than the developer hard-coding one effort setting for the whole session.
It installs in one line through pi's own package manager, backed by an MIT-licensed GitHub repo with the extension's full source.
What you get
- Full TypeScript source code under the MIT license
- The jevort pi extension, installable with pi install npm:jevort






