Recently I built an ensemble of synthetic persona agents for a client in advertising.
It turned out to be a much better brainstorming tool than a single chat, because the differences between the personas are the useful part.
So I wrote an open-source desktop version.
You compose a council of agents with name, model and system prompt.
Every message you send is broadcast to all of them in parallel, each answers in its own column.
It use your OpenRouter key, one request per agent.
The obvious cost is that N personas means N× the tokens, so I suggest you to keep the councils small and use cheap models if possible.
So I wrote an open-source desktop version. You compose a council of agents with name, model and system prompt. Every message you send is broadcast to all of them in parallel, each answers in its own column.
It use your OpenRouter key, one request per agent. The obvious cost is that N personas means N× the tokens, so I suggest you to keep the councils small and use cheap models if possible.
Electron + Bun + React, MIT license.
Happy to hear what you think.