What it does
It routes subagents without disturbing the main chat's prompt cache, and falls back to passthrough after repeated crashes.
Agent tooling
A local proxy that uses Jev to choose the Claude model and reasoning effort for each message.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It routes subagents without disturbing the main chat's prompt cache, and falls back to passthrough after repeated crashes.
Ask your developer to add this tool to Claude Code, your coding assistant. Set it to choose cheaper AI for separate tasks while keeping the main chat on the same AI. Switching during a long chat can cost more because the history must be read again. Check the dashboard on your computer to compare spending.
Your requests go to TypeSafe, an outside AI service, which chooses the AI for the task. TypeSafe does not write the answer. Make sure you are allowed to share your requests before using the tool.