How it works

1. A Button trigger takes the chat message in the text field. 2. Extract Information classifies it as general, reasoning, coding or search (request_type). 3. A switch on request_type picks the branch. 4. Coding goes to an AI Agent on Claude, reasoning to Gemini Pro, search to Gemini Flash, anything else to GPT-4.1 mini. 5. Every AI Agent has memory on, so the conversation continues across messages. 6. Connect the four model connections and set a per-user memory session id after installing.