Wenlan
Wenlan is a source-backed AI knowledge base and LLM wiki for AI work: organize documents and decisions into pages you can find, inspect, and review.
Гайдук-помічник, чати з експертами й наради кількох моделей (OpenAI, Claude, Gemini, OpenRouter, NVIDIA). Файли й ключі лишаються на пристрої.
Translation pending
ProductivityPythonGitHub @molfarsystemsOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
No updates detected yet. GitHub releases will appear here.
Role packs and ready-made roles for Molfar System — the Android app that runs meetings between several AI models. Install a whole pack in the app from one .zip. Free to copy and adapt.
Loading the full README.
Read the full README ↗English · Українська
Roles for Molfar System — a free Android app where several AI models hold a meeting instead of one chatbot answering alone. Here you'll find ready-made role packs that go into the app as a single file, and individual roles to copy by hand.
The app never contacts this repository on its own: you decide what to install.
A role is who the model will be in a conversation: the prompt (who it is and how it talks) and the skill (what it does and in what order), up to 25,000 characters each. Roles live in two catalogs in the app:
The app already comes with 29 built-in roles. This repository holds the ones it doesn't have.
A pack is a .zip file the app adds to your catalog. You need Molfar 2.0.5 or newer.
roles/packs/ (Molfar and community roles) or from external/ (selections from other open collections). On GitHub: open the file → Download raw file.Nothing is deleted and nothing restarts:
A pack doesn't touch your API keys, chats or meetings — it holds roles only.
roles/en/devils-advocate/.prompt-*.md → save. The same with skill-*.md in the Skill block.Settings → Data archive → Create archive → Roles only — to share. You get a .zip with your own and changed roles — send it to friends or colleagues: they install it exactly like a pack from here.
Want your role to be here? Open a New issue → the Submit a role form. No git needed: fill in the fields, submit, and the role appears here with a link to you.
What is a role? A prompt (who the model is and how it talks) plus a skill (what it does and in what order), up to 25,000 characters each.
What do I need? An Android phone with Android 7.0 or newer, Molfar 2.0.5 or newer to install packs, and your own key from an AI provider.
Will installing a pack delete my roles? No. Identical roles are skipped; for a role with the same name but different text you choose Replace, Keep both or Skip. Nothing is deleted.
Where do my requests go? Straight from your phone to the AI provider you chose. Molfar has no server of its own.
Can I contribute without git? Yes: New issue → "Submit a role". Fill in the form and a bot turns it into a pull request.
We are a solo developer from Ukraine, working with Claude. Replies may be slow — questions are welcome in Discussions, and this FAQ covers the common ones.
| Folder | What is inside |
|---|---|
roles/packs/ | Ready-made .zip packs — installed via "Data archive" |
roles/ | Individual Molfar and community roles: prompt, skill and description for each |
roles/builtin-ua/ | The 26 built-in roles in Ukrainian — texts and a pack |
external/ | Selections from other open role collections — under their own licenses |
external/agency-agents/ | 151 roles from The Agency (agency-agents), 12 packs by division — MIT |
examples/ | Real meetings, exported as they ran: a 4-week campaign planned by six models with The Agency's roles |
FORMAT.md, tools/ | The pack format and the scripts that build and check packs — make your own |
Languages. Roles are not mirrored: a role exists in the languages someone wrote it in. The language of a role doesn't set the language of the answer — the model answers in the language of the question.
Models. There are no model lists here: the app loads each provider's models itself over the API (AI Settings → provider card → Models → Load from API). The keys are yours; providers — Google Gemini, OpenAI, Anthropic Claude, OpenRouter, NVIDIA NIM or any OpenAI-compatible service.
Everything in this repository except the external/ folder and the model answers in examples/*/meeting/ is CC0 1.0, public domain: take it, change it, publish it, use it commercially. Credit is not required, but appreciated.
The contents of external/ are under their authors' licenses (each subfolder has its LICENSE and source). The files in examples/*/meeting/ are unedited model output: we do not claim copyright in them and do not dedicate them to the public domain.