Private AI for Android
Type a question, or tap the microphone and say it out loud. Either way Lite Mind works the answer out on the device, and nothing you say leaves it.
Free, no ads, nothing to sign up for. Android 8.0 and up.
Both of them stay on the phone.
Pick a model and download it once. After that the app works in aeroplane mode, on a train, anywhere. Replies arrive as the model writes them and the chat follows along. Your history is kept on the phone.
Tap the microphone, talk, tap stop. whisper.cpp turns the recording into text on the phone and drops it into the message box, where you can edit it before sending. Your voice never leaves the device either.
Both are small enough to sit on a phone. The app fetches them from Hugging Face, and you can switch between them whenever you like.
Answers in a second or two and runs comfortably on older phones. Good for everyday questions, short writing and quick explanations.
Handles harder questions and writes longer, more accurate answers. Built for phones, so it stays quick despite being the larger of the two.
The app opens straight into the chat screen. There is no account to make and no setup wizard to sit through.
Pick Fast or Smarter from the sheet that sits over the chat. It is a few hundred megabytes, fetched once over your connection.
From then on it works in aeroplane mode, on a train, or anywhere else you happen to be. The model is already on your phone.
There is nowhere for your conversation to go. The app has no server behind it, and no privacy setting you need to hunt for and switch on.
No accounts, no analytics, no crash reporting to anyone else. Your chats and settings live on the phone, and uninstalling takes them with it.
Fetching a model from Hugging Face over HTTPS, and asking Google Play whether a newer version exists. Never for answering you.
whisper.cpp transcribes on the device. No recording is uploaded, and the microphone is only asked for when you tap it.
Lite Mind is a hobby project, written by one person. A short note on what is doing the work underneath, for anyone who likes to know.
The two engines that do the work, compiled for the phone. Chat templates come from each model file's own metadata.
The app reads what each CPU core can do, keeps the fast ones and pins its worker threads to them. It also loads the quickest backend your phone supports.
Smarter ships with a small draft model that proposes tokens for the big one to check in a single pass. The reply comes out identical, just sooner.
Yes, once a model is downloaded. That download is the only part that needs the internet. After it finishes you can switch the connection off and carry on.
Android 8.0 or newer on a modern ARM phone, which covers nearly everything sold since about 2017. Fast wants 2 GB of RAM and Smarter 3 GB, plus room for the model file itself.
These are small models, roughly a thousandth the size of the ones behind the big cloud assistants. They do well on everyday questions, explanations and short writing. They also get things wrong, so check anything that matters.
Hugging Face. Qwen3.5-0.8B comes from the bartowski GGUF build and LFM2.5-1.2B from Liquid AI. Both are published under licences that allow this use.
No. There is no price, no subscription, no advertising and nothing to unlock. It is a personal project rather than a business.
Check it yourself. Once a model has downloaded, put the phone in aeroplane mode and keep chatting. The replies still come, because the model is sitting on the phone. Android's own data usage screen will back that up.
Free, no ads, nothing to sign up for. Download it once and it works anywhere.
Lite Mind