vbookshelf/Chat-With-Two-LLM-Agents-Python-Workflow — reverse-engineered prompt

Build me a simple local Python workflow, ideally as a Jupyter notebook and a plain script, where one person can chat with two AI agents at the same time. I want it to feel like a small panel discussion. The user should act as the moderator, type messages in a basic input, and then a router should decide whether the next speaker is the user, agent one, or agent two based on the conversation so far.

Use the example setup from the README, with one agent as a warm psychologist named Emma and the other as a witty historian named Liam, talking about the rise of virtual girlfriends. The two agents should each keep their own chat memory, and there should also be one shared master conversation history so everyone can stay in sync. Make the router return only the next speaker in clean JSON.

Please set it up to run locally with Ollama, using the Gemma 3 12B model if available. Keep it easy to run and easy to tweak for different expert roles or topics. Look up current docs online if needed.