Molfar System
Four AI models debate your question on an Android phone
Molfar System isn't a chat with one bot — it's a meeting of several AI participants on one screen. You seat up to four of them at the table, each on its own model and its own provider, with its own role and its own documents. They answer independently and can't see each other's replies, so no model adapts to an opinion already voiced. The moderator compares every answer, pulls them into a single summary, and builds a queue of follow-up questions for specific participants on its own. No backend, no accounts, no monetization: everything lives on the device and requests go straight to the provider with your own keys.
Features
— Up to four AI participants at once, each on its own model and provider (Google Gemini, NVIDIA NIM, OpenRouter)
— Participants can't see each other's replies — no echoing of the first opinion in the room
— The moderator compares all answers: shared points, disagreements, recommendations
— A follow-up queue the moderator drafts itself — edit the questions, redirect them, or run the round as it stands
— The secretary maintains the project's knowledge base, names the source file it quotes, and writes the meeting protocol
— Gaiduk: a single chat for quick questions, any role, up to five documents
— 23 ready-made roles with editable prompts, plus your own
— Any model works — paste its ID and it appears in the picker, no app update needed
— Documents in txt, md, docx and pdf, indexed and stored on the device only
— Long-term project memory: decisions, deadlines and agreements survive between sessions
— Chat export to markdown, readable in the app or saved to Downloads
— BYOK: your own API keys, all three providers have free tiers
— No backend, no accounts, no analytics, no ads, no in-app purchases
— Five interface languages, light and dark themes, Android 7.0 and up
Use Cases
— Lawyers, analysts and auditors working with documents that can't go to someone else's cloud
— Decisions where one answer isn't convincing: compare positions, see what a single model missed
— Reviewing your own documents through several expert roles instead of reading them through once
— Anyone curious about multi-model setups who doesn't want to wire up Python, a terminal and agent frameworks
— Working without subscriptions: the providers' free tiers cover everyday use
Comments
Seven months ago I wrote the first line of this app on my phone — and never once sat down at a computer. The app, the website, the images: all of it was made on a phone. The idea came out of a habit. I'd ask the same question in three different chat windows and compare the answers by hand. That's tedious, and worse — the moment you paste one model's answer into another window, it starts agreeing with it. So I moved the comparison inside the app and made sure the participants can't see each other. It's a semi-research project: no monetization, no investors, no company. I'm testing where several independent models genuinely produce a better result, and where the extra complexity isn't worth it. Part of it works, part still needs proving. Which means I'm not only after positive feedback. If you try it on a real task and the multi-model setup gives you nothing useful — tell me. That result is the most valuable one right now.
Premium Products
Sponsors
Buy
Makers
Makers
Comments
Seven months ago I wrote the first line of this app on my phone — and never once sat down at a computer. The app, the website, the images: all of it was made on a phone. The idea came out of a habit. I'd ask the same question in three different chat windows and compare the answers by hand. That's tedious, and worse — the moment you paste one model's answer into another window, it starts agreeing with it. So I moved the comparison inside the app and made sure the participants can't see each other. It's a semi-research project: no monetization, no investors, no company. I'm testing where several independent models genuinely produce a better result, and where the extra complexity isn't worth it. Part of it works, part still needs proving. Which means I'm not only after positive feedback. If you try it on a real task and the multi-model setup gives you nothing useful — tell me. That result is the most valuable one right now.
Premium Products
New to Fazier?
Find your next favorite product or submit your own. Made by @FalakDigital.
Copyright ©2025. All Rights Reserved