Jev in production › Model and agent routing
Local proxy for Claude Code and Codex sends each new task to Jev in one request for a capability tier and per-model reasoning effort with confidence; a user-editable policy maps the answers to a model and pins it for the conversation.

[!IMPORTANT] This project has been retired and is no longer maintained.
I built Switchboard as an experimental prototype to explore what System One models could do for AI routing. Through my testing and following developments in the community, I concluded that this approach wasn’t reliable enough for coding tools. A task’s complexity depends heavily on the surrounding code and conversation, and can be very different from what the prompt alone suggests.
For the project's own README, linking back here: