Download a model once. Air Infer handles chat, tools, voice, and phone actions entirely on-device — your prompts and conversations stay on your phone, not on our servers. No cloud LLM for inference, no API key required.
Chat
Draft a reply without sending my data anywhere
Here's a draft — generated entirely on-device.
Capabilities
A complete on-device assistant stack. Every feature runs locally — no internet needed after the initial model download.
Multi-turn conversations with full streaming output. Every thread is saved locally — rename, delete, or export any reply as a PDF. Auto-titles keep your history organised.
Browse and download models from within the app. Supports LiteRT-LM (.litertlm) for Google's optimised format and llama.rn (.gguf) for the broader open-source ecosystem.
Create reusable AI mini-tools through a simple 3-step wizard — no code required. Give your tool a name, describe what it does, and it uses whichever model you have loaded.
Turn plain-language commands into real Android actions. Air Infer converts text to structured JSON and fires Android Intents — no Accessibility Service required.
Whisper runs fully on-device via whisper.rn. Speak into Chat or Mobile Actions without a cloud speech API. Choose tiny, base, or small model sizes in Settings.
Photograph a document or pick from gallery — ML Kit OCR extracts the text and injects it into your chat context. Supports multiple scripts, stays completely on-device.
Running a long generation on a warm device? Enable Cool Mode to throttle inference speed and reduce heat output — keeping your phone comfortable during extended sessions.
Generate images entirely offline. Air Infer renders SVG-based images through an on-device pipeline — no internet, no external API, no image ever leaves your phone.
Simple flow
Open the download page on your Android phone, grab the package, and allow installs from your browser if prompted.
Download from the in-app catalog or paste a direct URL. Supports LiteRT-LM (.litertlm) and llama.rn (.gguf) formats.
Streaming replies, saved threads, auto-titles, PDF export. Text generation never leaves your phone.
Build mini-tools with the Tool Builder wizard, or use Mobile Actions to fire Android intents from plain language.
Once a model is loaded, text generation happens locally. There's no default cloud LLM completing your messages. Network is used only for downloading model files — not for inference.
Your feedback directly shapes what we fix and build next. Takes 2 minutes — no account needed.
Air Infer is invite-only while we polish things before Google Play. Android users can download the APK directly, or join the waitlist for priority access and build notifications.
Priority access + build notes when we open beyond the beta.
No spam. Unsubscribe any time.