Mrai Buddy

Mrai Buddy — free AI phone assistant (floating bubble, phone control, keyless fallback AI chain). Based on Buddy by ghorbelhamdi, Apache 2.0.

What it can do Uses your apps for you Messages, shopping, smart home, settings, games: anything on screen. Add instructions while it works, by typing or by voice. Glass everywhere Set any image — or a video — as your background and every screen (chats, settings, terminal, sessions) floats over it as tinted glass. The tint comes from the image itself, never a hardcoded colour. Each screen can carry its own background or follow the chat's. The Sharingan lives The bubble's eye plays rendered animation loops: a slow, expressive blink when idle, a narrowed smoulder when fired up. It watches, it waits, it blinks like it means it. Your system prompt is law Write your own system prompt in Settings and it rules every backend: it replaces the default persona on Mrai AI, Custom AI and Local AI, and leads the system prompt — above the app's own manual — on Claude, Codex and OpenCode. No length limit; long prompts collapse into a pill so they never eat the screen. Six AI backends Mrai AI (free, no key), Claude Code, Codex, OpenCode's free models, your own Custom AI endpoint, or Local AI running entirely on your phone. Local AI, fully offline A sub-500 MB model running on-device via llama.cpp. No account, no key, no network. Private by design. Private browser For news, prices and current information, the agent browses the web silently — no visible browser, no tracking. It answers with what it found. Approvals are your setting Action approvals (Settings → Experience) decide who gates a Send / Pay / Delete: with them on, you get Approve / Deny / Change ("make it 9, add a smiley") cards; with them off (the default) the agent just does what you asked — no popups, ever. Structural protections stay on either way. Asks when it's unsure Questions come as a card with answer buttons, or type your own answer. Voice Talk to it with the phone's recognizer, Claude's or ChatGPT's dictation, and hear replies in ChatGPT's voices. Plain mode No moralizing, no lectures. Your instructions, your phone. Updates itself A notice when a new version is out, or just ask: "update yourself".