What it does
A YAML rulebook narrows the candidates first, then Jev answers one question — which handler owns this — and the verdict routes, blocks, or asks you.
Agent tooling
Watches what you type into Claude Code and decides who should handle it: your slash commands go straight through, everything else gets checked against your rules.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
A YAML rulebook narrows the candidates first, then Jev answers one question — which handler owns this — and the verdict routes, blocks, or asks you.
You can use this tool to direct requests in your own AI assistant. A developer will first set up an access key. This key connects your software to an outside AI service called TypeSafe. When someone types a prompt, your app sends the text to TypeSafe. That service then decides which of your specific tools should answer.
Your developer can start by running the included marketing team example in shadow mode. This mode safely records the decisions the tool would make without actually changing anything. Note that the tool sends up to 6000 characters of each typed prompt to TypeSafe. It tries to hide passwords before sending, but this is not guaranteed to catch everything.