Jev in production › Model and agent routing
pi-pignon
Pi coding-agent extension uses local or hosted Jev to judge prompt difficulty and pick the LLM tier from a user-defined routing table.
The source publishes no measured number.
- Use
- Model and agent routing
- Industry
- Developer tools
- Form
- Open-source tool
- Stage
- Announced
- Plugs into
- Pi
- Listed
- 2026-09-25
- Found via
- github
- Repository
- siiick/pi-pignon
- Stars
- 5
- Forks
- 0
- Last push
- 2026-10-08
- Language
- TypeScript
- License
- MIT
The README opens with
Pi agent extension that shifts to the right LLM for each prompt, the way a bike changes sprocket (pignon): a small decision model judges how hard the prompt is, and pignon looks the answer up in your routing table.
- Decisions come from a local Laya System-1 model (served by laya-serve, ~75 ms on Apple Silicon) or from TypeSafe's hosted Jev (~70–500 ms, needs an API key) - You choose the models and write your own difficulty tiers - Strongly typed TypeScript, config checked against a published JSON Schema
Badge
For the project's own README, linking back here:
Also used for model and agent routing
- Hermes Jev Skills ~0.4 s per turn
- Jev Codex Router ≈ −60 %
- Quicksilver 86%
- Grok Bot Jev Router 13.0×
- Jevstiller 4,000
- jev-use ~230 ms
- DuoMind 36%
- jevcache (kushals256) 394ms