(and maybe something else!)
You MUST find each and every way to make mistakes while still adhering to the prompt
- 100% Offline & Private by design: Inference runs on-device with Gemini nano (
Google AI Core). Prompts, chats, and attachments stay on your phone (optional online extensions are strictly opt-in and off by default). - On-Device Models, Multi-Model Manager & Thinking Mode:
- Switch between Stable (Currently
gemini-nano-v3) and Preview/Beta (gemini-nano-v4-full/v4-fast) channels. - Download any model your device supports.
- Thinking mode with Nano-v4.
- Custom
[USER INSTRUCTIONS]Markdown prompt import/export, per-chat temperature & max-token controls, optional time/language context, and auto-continuation for longer answers.
- Switch between Stable (Currently
- System Assistant Role, Vision & Screen Context:
- Set PAIOS as Android's Digital Assistant!
- Capture photo or attach from the gallery (yes, it's been a year, it is finally done).
- Intelligent ability to check out the screen context in assistant mode - when needed.
- RAG, Agentism & Modular Extensions:
- Using IPC (already implemented in PTK, and upcoming
Air Raid Alerts) supporting up to 6-turn tool chaining and per-app / per-module Extensions Settings. - Tool calls as bundled modules (Documentation, basic Device Info and screen resolver) and opt-in modules for online capabilities (Wikipedia, DuckDuckGo, Open-Meteo Weather).
- Using IPC (already implemented in PTK, and upcoming
- M3 Expressive UI, Docs Explorer, 5 Languages and so much more:
- Custom UI package (TBD) that builds upon Material 3 eXpressive
- 23-page GenAI/AICore Docs Explorer (bundled offline + live
.md.txtsync + RAG tool) - Multilingual support - 5 languages (
EN / UK / ZH / DE / TR- or you can add yours :) ), - GitHub issue and documentation viewer in-app
- Changelog and package list viewer in-app
- Transparent System Prompt and prompt viewer in-app.
Google AI Core / ML Kit GenAI Prompt API is now in Beta, powering gemini-nano-v3 (Stable) and gemini-nano-v4 (Preview/Beta) on-device.
- Device & OS Requirements: Requires Android 14+ (
minSdk 34), 8 GB+ RAM, a locked bootloader, and Google AI Core across supported chipsets including Pixel 8/9/10 (Tensor G3/G4+), Samsung Galaxy S24/Z6/S25 (Snapdragon 8 Gen 3/Elite, Exynos 2400/2500), and devices powered by MediaTek Dimensity 9300+. - Model & Safety Behavior: Native on-device limits (256-token output cap / 25s generation window) are mitigated by bounded auto-continuation and cutoff rescue, though compact on-device reasoning remains smaller than cloud LLMs (such as Gemini Pro/Ultra). AICore safety/policy filters trigger automatic
[POLICY FALLBACK RETRY]or clean error states rather than an unrecoverable soft-lock.
- Google Group Tester Opt-In: Opt in via Google Group and then joining via Play Store
- Google Play Open Testing: Enroll on the app's Google Play Store page
- GitHub Nightly Prereleases: Puzzaks/PAIOS Releases
- FAQ: Frequently Asked Questions
- Support & Help: Support Guide
- Changelog: Full history
- Roadmap: Future plans
- Contributing & Translations: How to build and help
- Security: Security Policy
- Code of Conduct: Code of Conduct
This is an independent, third-party app and is not officially affiliated with, endorsed by, or sponsored by Google. "Gemini" is a trademark of Google, LLC.
Note: Versions prior to 1.1.2 used a legacy package ID and are no longer supported — see the FAQ.
Huge thanks to @kekko7072 for inspiring this project with this plugin!
Google Antigravity for AI reviews and suggestions during development!
And to Armed forces of Ukraine for keeping me safe. Stand with Ukraine!