Choose the device for the work you want to finish.
Off Grid AI has mobile and desktop apps, but their features are not identical. OGAM (Off Grid AI Mobile) brings local AI to supported Android phones and iPhones. OGAD (Off Grid AI Desktop) adds a desktop workspace, with feature-specific Pro tools. Linux core support is available in beta108; that release does not include Linux Pro.
| Find your download | Desktop beta108 release |
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.
What can you use across phones and computers?
Local text chat is a common starting point. Download a compatible model, give it a bounded task and review its answer. Image understanding, image generation and speech need their own compatible models or resources. A model that appears in one catalog is not proof that its files and acceleration backend work on every platform.
This guide reflects the releases checked on September 29, 2026. Use the installed version’s model choices and feature gates when following it.
| Device | Useful starting point | Important distinction |
|---|---|---|
| Supported Android phone | Local chat, supported image workflows and voice input | GPU/NPU acceleration depends on the phone and model |
| Supported iPhone | Local chat, supported image workflows and voice input | Use the iOS-compatible model options |
| Apple Silicon Mac | Core AI workspace and supported Pro work tools | Background capture requires opt-in and permissions |
| Windows x64 PC | Core AI, plus supported Windows Pro features | Do not assume every voice or native workflow matches Mac |
| Ubuntu x64 computer | Linux core in beta108 | This beta does not bundle Pro |
The current Linux release targets Ubuntu 24.04 or newer on x64. It does not provide a Linux ARM package. Windows and Linux optional CUDA packs belong to the beta setup; they are not a universal requirement for local chat. Linux and GPU release.
Which device should you use for writing and documents?
Use the device where the source material is easiest to review. A phone can help rewrite a short note while you are away from your desk. A desktop is often more convenient for reading longer source files and comparing an answer with the original.
OGAD’s core Projects workspace lets you work with supported uploaded documents and project conversations. Start with one small document whose contents you know. Ask a specific question and inspect the cited source rather than accepting a fluent summary on its own.
For example:
From this project brief, list the deliverables, the approval step and any stated deadline. Quote the relevant source for each. Say when a detail is absent.
The useful result is a checked brief you can work from. It is not a claim that the app has read every file on the computer or can infer missing terms.
What differs for images?
A vision model answers questions about an input image. An image-generation model creates a new image. Those are separate capabilities, even when both appear in the same app.
On a phone, choose the model offered for your platform and hardware. Android and iOS image runtimes differ, so do not transfer expectations from one device’s results to another. On desktop, check the selected image model and active backend before attempting a large image.
Try a simple request first: a clean illustration of a houseplant beside a window, with no text. Once it works, change one element at a time. Keep the first result as a comparison if you change model, dimensions or settings.
A successful text answer does not establish that image generation is prepared. Complete the separate model download before expecting it to work offline.
What differs for voice?
There are at least three distinct voice tasks: dictate into a chat, hear a spoken reply, and record a meeting or dictate into another app. They do not share one universal feature gate.
| Voice task | What to check |
|---|---|
| Speak into a chat | Local transcription model and microphone permission |
| Hear the answer | Speech model, voice resources and platform/tier support |
| Dictate at another app’s cursor | Desktop Pro Voice workflow and required permissions |
| Record and transcribe a meeting | Supported Pro meeting workflow and audio inputs |
Mobile spoken output has Pro requirements; it is not the same as free chat dictation. Windows beta108 supplies local English US/UK speech voices through its ONNX route. That does not establish the same multilingual voice catalog as Mac or mobile.
Linux core voice support also does not establish Linux Pro meeting recording. Use the task and the tier together when choosing a device.
Where do background work history and automatic to-dos fit?
These are supported desktop Pro workflows. They depend on retained inputs and setup, rather than simply opening a chat model. Capture is opt-in with a visible recording state. Screen history is sampled activity, not a complete continuous video of everything you did.
Windows has shipped Pro-capable builds, including verified Vault, Clipboard, Replay and Devices routes. A blanket statement that Windows has no Pro is therefore wrong. At the same time, checking those features does not prove every Mac-specific audio or accessibility route behaves identically on Windows.
If recovering past work is your goal, test that exact path: enable the required capture, create a harmless sample, confirm it appears and find it again. A menu item alone is not a completed setup.
Can you continue work on another device?
Supported device sync moves selected data between your paired devices. Check the data type, permissions and sync controls. Pairing two devices does not mean every file, model and private record should move automatically.
A local network can support local device communication without internet, once setup is ready. Reaching home hardware from elsewhere needs a usable network route. Tailscale/private-address access does not remove that connectivity requirement.
Also separate sync from remote inference. Sync transfers data; remote inference sends a request to a model running on another host. Choose the operation you actually need.
How do you choose without setting up everything?
Write down one outcome: draft a note on your phone, search a document on your computer, or recover a detail from captured work. Then check four things: device support, required model, free or Pro access, and network needs.
Run a small end-to-end example on that device before expanding. This is a practical way to learn the limits without downloading a model library you will not use.
Get Off Grid AI for your device and start with that one result. Add another capability only when it helps you finish the next task.