One question, several AI models: parallel answers, debate to consensus, or a judged verdict. Free, open-sourced and browser-only β no accounts, no server; keys and history live only in your browser, prompts go only to the providers you choose.
Hi Product Hunt β I built this, happy to answer anything.
The origin is a ritual you might share: any time a question actually mattered, I'd paste it into ChatGPT, Claude, and Gemini in three tabs and eyeball the differences. The comparison was the value, and it all happened in my head.
Yes-Brainer turns that into a first-class thing: a council. One composer, several models seated (Claude, GPT, Gemini, Groq, OpenRouter, local Ollama β any mix). Three ways to deliberate: parallel answers you compare yourself; a trial where models rate each other's answers anonymously and a judge writes one verdict; and a consensus debate, where they re-answer across rounds against each other's arguments until a mediator sees genuine convergence β or honestly reports what's still contested.
Two things I want to be upfront about:
Your keys. The app has no server β it's a static bundle. Keys live in your browser and go straight to the providers; history stays in your browser too. Please still do your part: use a dedicated key with a spending cap, and check the address bar says yesbrainer.ai before pasting. The code is open source (AGPL-3.0), so the key-handling is readable, not a promise.
The limits. More models isn't truth. A council reduces single-model blind spots, but it can still be confidently wrong together β the app says this on every thread. It shows you the spread between models; the judgment stays yours.
You can read recorded demo councils on the site without pasting any key. It's free β no subscription, no paywall, no accounts. The concept credit goes to karpathy/llm-council; this is the production-shaped take β three structures, streaming, persistence, mobile PWA, zero backend.
Feedback, ideas, and improvements are welcome.
Report
ran a coding prompt through it and the consensus round actually picked a better answer than either model gave solo, which kind of surprised me. love that nothing leaves the browser too.
Report
Tried it with a coding question and watching the models argue it out before the judged verdict picked a winner was honestly pretty cool. Loving that nothing leaves my browser.
Ok! So...
ran a coding prompt through it and the consensus round actually picked a better answer than either model gave solo, which kind of surprised me. love that nothing leaves the browser too.
Tried it with a coding question and watching the models argue it out before the judged verdict picked a winner was honestly pretty cool. Loving that nothing leaves my browser.