Offline Local AI Assistant Phone App (LLM Hub)

Overview

LLM Hub is a private AI assistant app for Android and iOS. Everything runs on the phone: local LLM chat with Gemma, Llama and other GGUF models, Stable Diffusion images, Whisper transcription, translation, a scam detector and a small vibe-coding agent. No account, no servers, and prompts never leave the device during local inference.

Melbourne developer Timmy Qian (listed on Google Play as Yuan Qian) built it as a passion project, open-sourced it on GitHub, and then turned on monetization through a premium tier and ads. Because users’ phones do the computing, he says the business has $0 server costs.

This is about as pure a local-AI stack as it gets. The product’s value is the packaging: compressing open models, tuning them for phone CPUs, GPUs and NPUs, and wrapping them in an app normal people can use offline.

Realistic income

These figures were shared by the founder. On Mar 22, 2026, Timmy Qian posted that LLM Hub made A$823 gross in the 10 days after he enabled monetization, with 25,000 installs. On Apr 12, 2026, he posted month-one revenue of A$3,234 (about $2,284 USD) gross. The Appfigures dashboard on his studio site shows A$13,726 all-time gross for LLM Hub and A$1,532 in the last 30 days (seen Oct 24, 2026). So recent monthly revenue is below the month-one figure.

These are gross store figures before store fees, from one solo developer. They don’t predict your results.

How to build your own

• Start from an open-source runtime: llama.cpp for chat and whisper.cpp for transcription. Both are free.
• Choose a niche the general apps ignore, such as an offline assistant for travelers, field workers, a specific language, or privacy-sensitive jobs. Then pick small models (1-4B, quantized) that run on mid-range phones.
• Ship on Android first. The Play Store is the cheaper and faster place to test demand, as the operator found.
• Make money with a one-time premium unlock or subscription plus optional ads. RevenueCat is free until $2,500 in monthly tracked revenue.
• Use AI coding tools to move fast, the way the operator vibe-coded the app. Still test on real devices, because crashes and model-download bugs show up in reviews (the app sits at 3.2 stars on Google Play).

Risks

• Revenue is self-reported and has slowed since the first month (A$1,532 in the last 30 days vs A$3,234 in month 1).
• Free competitors and built-in phone AI (Apple Intelligence, Gemini Nano) may squeeze paid local-AI apps.
• Many devices and chipsets mean constant crash and memory bugs, which hurt ratings.
• Open model licenses (Gemma, Llama, Stable Diffusion) each have their own terms you must follow.
• App store fees and policy changes cut into gross revenue.