Get the Saient Android app
Saient is published on Google Play as Saient: Offline AI & Video, package uk.co.saient.app. It is free and rated PEGI 3.
What the app does on your phone
Local AI chat
A quantized GGUF language model answers you on-device. No account, no API key, no message leaving the phone. Chat history is kept in the app's private storage and can be reloaded or cleared.
Image generation
Genuine SDXL running through Quartz — model parsing, tokenization, scheduling, operators, memory management and Vulkan dispatch are all Saient's own code. Generate at 512, 768 or 1024 pixels, with fast 4/8-step Lightning profiles.
Text-to-video
Wan2.1 T2V 1.3B generating real clips on the phone's GPU. Pick a step count and clip length, then generate locally. See how Wan2.1 runs on Android.
Assistant tools
Ask it to look up a contact, open the dialer, or draft a message for you to send yourself — and search the web when you ask for it. You stay in the loop for anything that leaves the device.
Saient running on an Android phone
Why "offline" here is a fact, not a marketing word
There is no inference server behind this app. The engine is compiled into the application package and runs as a process on your phone, so there is nothing for a prompt to be sent to. The check is simple and you can run it yourself: put the phone in aeroplane mode and generate.
The only things that use the network are the one-time model downloads, any web lookup you explicitly ask the assistant to do, links you choose to open, and optional safety reports. A safety report contains a category, the feature and model involved, and the app version — never your prompt, your chat or your generated file.
Full detail is in the Saient mobile privacy policy.
What your phone needs
| Requirement | Detail |
|---|---|
| Platform | Android. Package uk.co.saient.app, distributed through Google Play. |
| GPU | A Vulkan-capable GPU for image and video generation. Chat runs on more modest devices. |
| Memory | Around 8GB RAM is the target class for generation work. |
| Storage | Depends on which packs you install — about 2.93GB for the Wan2.1 video pack (allow roughly 4.7GB free while it installs), about 6.94GB for the SDXL FP16 image pack, and around 1.3GB for a Qwen3 1.7B chat model. Install only what you want. |
| Network | Needed for the initial model download, optional web lookups, and optional safety reports. Not needed for generation. |
| Account | None. No sign-in, no licence key, no API key. |
| Verified on | Samsung Galaxy S24 (SM-S921B). |
Quartz — the runtime inside the app
Saient mobile does not wrap llama.cpp, ONNX Runtime, PyTorch or a hosted API. It ships Quartz, Saient's from-scratch Rust inference engine, built for ARM64 and driving the phone's GPU through hand-written Vulkan compute shaders. Quartz is open source under the MIT licence at github.com/SaientAI/saient-quartz.
If you want the mechanism rather than the summary, the on-device video pipeline is written up in detail at how Wan2.1 runs on a phone.
The rest of the local-first stack
- Saient for desktop — the full local AI workshop for Linux and Windows: chat, a coding agent, SDXL images, Wan video, text-to-speech and LoRA training.
- The Saient agent — files, terminal, planner, memory and checkpoints, running against a local model.
- Saient compared with ComfyUI — where each one fits.
- Quartz — the open-source engine shared by the desktop and mobile apps.
- Agent persistence research — the evidence-first paper behind the agent's memory and continuity work.
Saient Android app — questions
Does the app really work offline?
Yes. After the one-time model download, chat, images and video all run with Wi-Fi and mobile data switched off. There is no inference server in the loop.
How much does it cost?
It is free on Google Play. There is no hosted inference to pay for, and no account to create.
What phone do I need?
An Android phone with a Vulkan-capable GPU and roughly 8GB of RAM for image and video generation; chat runs on less. Verified on a Samsung Galaxy S24.
How long does video generation take?
Minutes, not seconds, and the phone will get warm. That is the honest cost of not sending your prompt to someone else's server.
Do my prompts or generated files leave the device?
No. They stay in the app's private storage. See the mobile privacy policy for the full list of what does use the network.
Is it just a front end for a cloud service?
No — the generation path makes no API call. The engine is Quartz, compiled into the app and running on your phone.
Is there an iPhone version?
Not yet. Saient is on Android via Google Play, and there is a separate desktop app for Linux and Windows.
Why are the models downloaded separately?
They are gigabytes in size, far more than an app package can carry. Saient downloads only the packs you pick, SHA-256 checks every file, resumes interrupted downloads and installs atomically.
Run AI on your own phone.
Free, no account, and nothing to send anywhere.