Agent tooling
Agent Squad's JevClassifier chooses the next specialist agent by comparing agent descriptions with a message and its conversation history.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The TypeScript and Python integrations put agent descriptions into Jev Choice options and return the winner with confidence. They require TypeSafe access, with context and rate limits still applying. Agent Squad's separate on-device Swift features do not make this Jev request local: it goes to the hosted service.
Describe each specialist's job clearly. JevClassifier uses those descriptions, the incoming message and relevant history to choose a destination. Keep the history for follow-up questions. Choosing the agent is only the routing step; that agent still has to do the requested work.