Sixteen free add-ons. No key, no sign-up and no cost, and most of them work with no internet at all. Here is each one: what it does, what it needs and its honest limits.
Free AI inside, in 20 seconds.
In the app: Settings, then Free AI helper. Green means ready on your phone. Tap a blue one to set it up.
Download once on Wi-Fi. After that they run on your phone, and nothing leaves it.
A small open model, Qwen 3.5, answers you right on the phone.
No key, no cost, nothing leaves the phone. Simpler than the big online AIs, but always there. It steps in by itself when you lose signal or hit a limit, and marks its answers so you know.
Describe a photo, read a sign or check homework, all offline.
Its eyes are a one-time download next to the model: 205 MB for Fast, 668 MB for Smarter.
The phone check reads your memory and chip, then suggests the model that stays quick.
It runs by itself the first time you open the app and takes about two seconds.
The phone AI uses the right number of processor threads for your chip.
Set from the phone check. A speed-up on the graphics chip of newer phones comes later.
Talk to the app with no internet, using your phone's own offline speech.
Uses your phone's language pack from Google, fetched on Wi-Fi. Slightly less accurate than online speech. Not every phone offers it.
Replies are read aloud with voices stored on the phone.
Voices marked offline in the voice picker need no internet. Your phone settings can download more for free.
Translate text, menus and conversations with language packs on the phone (Google ML Kit).
About 30 MB per language. Use the packs only when offline, always, or never. Always keeps every translation on your phone. The wording is a little plainer than the AI's, but instant and private.
Timers, alarms, reminders, calls and maps from the free AI, offline.
Works with Qwen 3.5 and Gemma 4. Ask "set a timer for 10 minutes" the way you would ask your online AI.
Finds your memory notes by meaning, so the assistant uses the right ones.
A 278 MB model (EmbeddingGemma) on the phone. Nothing is sent anywhere.
Chat with files reads the words off every page on the phone when your AI cannot read PDFs.
Works for the phone AI and other open models. Scanned PDFs too, when the text is printed.
With six pieces ready, the whole assistant works on a plane.
Phone AI, eyes, voice typing, voices, translator and memory search. The Free AI helper shows how many you have.
Bigger models for bigger phones, or the one built into your phone.
A newer small model: smarter writing and 140+ languages, on the phone.
A 3 GB download for phones with 8 GB of memory or more. On a smaller phone, Qwen 3.5 Smarter is the better pick.
Qwen 3.5 4B, Gemma 3 1B, Phi-4 mini, or any .gguf file you bring.
Every model shows what it needs and what it can do before you download it. Delete any one to free space.
A model built into a few newer Pixel and Samsung phones.
Downloaded through Google once, then it answers offline. Text only. The Free AI helper tells you if your phone has it.
No key and no cost, but your words leave the phone. You turn these on yourself.
Your computer runs Ollama or LM Studio, and your phone talks to it over Wi-Fi or Tailscale.
Big models, zero cost, your own hardware. Your words go only to your own computer.
A shared free service answers and draws small pictures, with no key.
Honest limits: it is shared by everyone, so it can be slow or full at busy times. Pictures are small, with a tiny watermark, about one a minute. Only the words you type go to their server, never your name, memory notes, photos or files, so don't type anything private. The app asks you to agree first.
Drawing pictures on the phone itself.
Today's picture models need more memory than phones have, and a picture would take minutes and look rough. It comes when a small one is good enough. Until then, use Image studio with a Gemini key, or the free cloud.
Google gives you a free Gemini key. It takes about two minutes, with no card. On the free tier, Google may use your chats to improve its products, so the app tells you before you start.
Every model is free and runs on the phone. Bigger ones answer better but need more memory.
How much memory (RAM) does your phone have? Look in Settings, then About phone. It is the RAM number, not storage.
Fast (Qwen 3.5 0.8B) is made for phones with 4 GB, so on yours it will take its time. Keep your online AI for everyday questions and let the phone AI cover the moments with no signal.
Likely mode Lite: still pictures and light effects, so the app stays smooth.
Add its eyes for 205 MB more if you want it to look at photos offline. Gemma 3 1B also fits, for text.
Likely mode Usually Balanced. Lite on phones with 4 cores or fewer.
Its eyes are 668 MB more. Keep about 1.7 GB free on your phone for it.
Likely mode Balanced: smooth moving pictures, with the heaviest effects off.
Its eyes are 986 MB more. Keep about 4.5 GB free. Qwen 3.5 4B and Phi-4 mini fit too.
Likely mode Full on a fast chip: every effect on. Otherwise Balanced.
A guide only. In the app, the phone check reads your real memory, chip and free space, and marks the best model for your phone.
| Model | Good for | Phone memory | Download | Sees photos | Phone actions |
|---|---|---|---|---|---|
| FastQwen 3.5 0.8B | Quick answers. Works on most phones. | 4 GB | 507 MB | Yes, +205 MB | Yes |
| SmarterQwen 3.5 2B | Better answers, a bit slower. | 6 GB | 1.2 GB | Yes, +668 MB | Yes |
| Gemma 4Gemma 4 E2B | 140+ languages, great at writing. | 8 GB | 3.0 GB | Yes, +986 MB | Yes |
| Qwen 3.5 4BModel shop | The smartest Qwen that fits a phone. | 8 GB | 2.6 GB | Yes, +672 MB | Yes |
| Gemma 3 1BModel shop | Small and quick, good for short notes. | 4 GB | 722 MB | No, text only | No |
| Phi-4 miniModel shop | Strong at maths and reasoning. | 8 GB | 2.5 GB | No, text only | No |
| Memory searchEmbeddingGemma 300M | Finds your notes by meaning. | Any | 278 MB | Not needed | Not needed |
| Your own modelAny .gguf file | Any small open model you trust. | Depends | Depends | Depends | Depends |
Phone memory means RAM. Sizes are the one-time download. Fast, Smarter and Gemma 4 are in the main list; the others are in the Model shop.
The app reads what your phone can do, then runs in a mode that stays smooth on it. No lag on a cheap phone, every effect on a fast one.
Auto follows the check. Change it any time in Settings, Phone check.
Phone AI, eyes, memory search, PDFs, offline voice typing, voices, translator packs, Gemini Nano and the phone check. Nothing is sent anywhere.
The models come from a public model library. Translator packs, offline speech and Gemini Nano come through Google. Only the file comes down; nothing about you goes up.
With My laptop AI, your words go to the computer you set up, over your Wi-Fi or Tailscale.
With the free cloud (beta), only the words you type go to the shared service's server, under its own terms, with no account or key. Never your name, memory notes, photos or files. The app asks you to agree first.
More in the privacy policy.
Qwen and Phi-4 mini are open models under the Apache 2.0 and MIT licences, and Gemma comes with its own terms. llama.cpp: MIT. The free cloud is a shared free service with its own terms, and the app names it before you start. Model names belong to their owners. Homeray is not made or endorsed by any AI company.
Ask for your early access spot. Every tool and every free AI is free to use.
Quick question first? Message me on WhatsApp
Auto chose:
Homeray