Show and tell: four Agency roles as a meeting of different AI models, on an Android phone

[molfarsystems]

Hi! I'm a solo developer from Ukraine, working with Claude, Anthropic's AI. We built Molfar System, a free Android app where several AI models hold a meeting on one question. We used four of your marketing roles that way, and I'd like to share how it went and ask two questions.

Two questions first, only if they suit you:

1. Translations. Molfar's interface is in English, Ukrainian, German, French and Spanish, and we would like to translate some of your roles into those languages. The translations would live in our own repository, not in yours, marked as translations, with the same attribution and a link to the English original. When they are ready, we'd open an issue for your translations table, as your maintainers suggest. We are not asking for an endorsement, only whether you're comfortable with this.

2. If you think the exercise below belongs in examples/, tell us and we'll open a pull request with one short file. A link is just as welcome, and so is nothing.

What we did: we took your "Marketing Campaign Launch" scenario and ran it as a real meeting. Content Creator, Twitter Engager, Reddit Community Builder and Analytics Reporter each sat on a different model; Molfar's own orchestrator joined their answers and asked follow-up questions; a secretary wrote the minutes. The task was real: a 4-week launch for our role kit, with zero budget and about 4 hours a week of one person's time. It took round 1, two rounds of follow-up questions and about $1.5. The result is one 16-step plan where every step has an owner, a done test and a measure. The orchestrator refused claims the facts did not support, and the agents found real problems on our own website.

The exports are published unedited, with the documents the models received and an honest "what did not go perfectly" section:
https://github.com/molfarsystems/molfar-system-kit/tree/main/examples/agency-campaign-launch
Video, recorded on the phone: https://youtu.be/ArGtVTWNf30

Your roles were easy to adapt: the Persona / Operations grouping in your CONTRIBUTING.md maps onto a Molfar role, which has a prompt (who the model is and how it speaks) and a skill (what it does and in what order). 151 agents are available to our users as installable packs, with an attribution line in every role and the full MIT license inside every pack.

About the app, honestly: it is new and almost nobody knows it yet. It is free, with no ads and no accounts. It has no server of its own: requests go from the phone straight to the provider the user chose, with the user's own key. The app itself has no trackers or analytics.

Roles adapted from The Agency (agency-agents), MIT License, © AgentLand Contributors. Molfar is not affiliated with, endorsed by, or a partner of The Agency or AgentLand. This post was written with Claude's help.

If you'd prefer a different attribution line, or want some agents left out, just tell me and I'll change it. Thank you for making the collection open.

— Molfar System (a solo developer from Ukraine, working with Claude)
https://molfar.nova-hata.com/?lang=en · https://play.google.com/store/apps/details?id=com.nova_hata.molfar