Roles Witch turns any OpenAI-compatible model into a set of assistants you define yourself.
Create a role, give it a name and a prompt, and start talking. Switch between roles whenever you like — each one keeps its own conversation, its own settings and its own model.
BRING YOUR OWN MODEL
There is no subscription here and no server of ours in the middle. Connect Roles Witch to any OpenAI-compatible endpoint using your own API key — a cloud provider, or a model running on your own machine over your local network. Pick a different model per role if you want.
TALK INSTEAD OF TYPING
Start a voice conversation and just speak. The assistant replies out loud, and the conversation keeps going while your phone is in your pocket or the screen is off.
SPEECH THAT STAYS ON YOUR PHONE
Offline speech recognition, powered by the open-source Vosk engine, runs entirely on your device — no audio is sent anywhere. English is built in; other languages download only if you choose them.
ROLES YOU CAN SHARE
Build a role once, then share it as a QR code. Scanning it imports the role, prompt and all.
YOUR DATA STAYS YOURS
No account. No sign-up. No ads, no analytics, no tracking of any kind. Conversations, roles and settings live on your device, and you can clear your history or reset the app at any time. Your messages go only to the endpoint you configured.
ALSO INCLUDED
• Markdown formatting in replies
• Attach photos for models that support vision
• Favourite messages and jump back to them
• Per-role model, voice and language settings
• English and Russian interface
Roles Witch is a client, not a model. It needs an API key from an OpenAI-compatible provider, or your own local server — model access is not included.
Bring your own model. Custom AI roles, voice chat and on-device speech.