Off Grid AI can run local AI tasks without internet after the required setup. OGAD (Off Grid AI Desktop) runs supported models on your computer, and OGAM (Off Grid AI Mobile) runs supported models on your phone. Local text, speech, image, and document workflows depend on having the right models and files ready. Downloads, online services, and remote-device access have different connection requirements.
Download OGAD | Desktop releases
What would you like to do with Off Grid AI?
Want help with this workflow, or a feature you would like us to add? Email support@offgridmobileai.co with your devices and the task you want to improve.
The useful rule: follow the whole task
“Offline” describes a complete path from input to result, not just the model. A text model may run locally while the document you need is still in a cloud drive. A local draft may be ready, while sending it to someone still needs internet.
Start by asking where the source is stored, where the model runs, and where the result must go.
Suppose you want to prepare a client update during a connection outage. If the source notes and text model are on your laptop, drafting can work locally. Checking a new email or sending the finished update remains a separate online step.
What can work on one prepared device?
The following are local workflow categories. Exact model availability and feature support depend on the device and installed release.
| Task | What must be ready locally |
|---|---|
| Text chat and drafting | Compatible text model and the input |
| Document questions | Readable files, text model, and local search resources |
| Speech-to-text | Supported local transcription model and audio input |
| Image generation | Compatible image model and its required files |
| Image understanding | A suitable vision-capable model and the image |
| Spoken output | Supported local speech-output resources |
A text-model download does not prepare all the other rows. Select the local model for each task you intend to use.
Do not assume that the same model files or language options apply to every operating system. Check the relevant model catalog and release support.
What needs internet during setup?
You normally need a connection to obtain the app, download models and runtimes, install updates, and retrieve files that are stored elsewhere. Optional paid features can have activation or setup requirements too.
Complete a real first task while connected. For desktop document work, create a project, add a readable source, wait for indexing, and ask a question with a known answer. This helps prepare local search resources as well as the text model.
A partially downloaded model or cloud placeholder is not ready for offline use. Check that the complete files are stored on the device.
How does offline document work behave?
In OGAD, select a local model under Models > Text. Use Projects > New project, then Knowledge & settings > Knowledge base > Add files. After indexing, open Chats > New chat inside the project.
The app retrieves relevant text passages and supplies them to the model. The desktop source shows the project and knowledge controls.
Use text-based PDF, DOCX, TXT, or Markdown sources for the text workflow. Scans and diagrams may need another prepared input path. Retrieval does not guarantee that every page appears in each answer, so check important claims against the original.
Core Projects and chat do not require enabling background capture.
What is different on a phone?
OGAM can use downloaded on-device models for supported tasks. For speech input, open Model Settings > Transcription (Speech to Text) > Transcription model and choose an on-device model. In Chat mode, hold the app microphone, speak, and release it to review the transcript in the composer.
A remote transcription selection changes the processing path. A local text model is a separate requirement if you want an offline AI answer after dictation.
Check OGAM's current device requirements and prepare the models before disconnecting. The phone's storage and memory limit which setup is practical.
Does local device communication need internet?
Not always. Two devices on a suitable local network can communicate without internet access. But they still need that local connection.
This distinction matters for device sync and for using a model hosted on another computer. Sync transfers selected content. Remote inference sends a request to the host running the model. They are different workflows with different settings.
If you use Off Grid AI sync, check the supported versions, pairing, tier, and selected sharing options. A fully disconnected device cannot reach another device merely because both are yours.
What about using a home model while away?
Accessing a home computer from another location ordinarily needs internet at both ends, even when you use private addresses through Tailscale. The host also needs to remain running and reachable.
A private network can provide an encrypted path, including a direct connection or relay depending on conditions. Private addressing does not make the remote task internet-free.
When the model runs on your home computer, the prompt leaves the client device and reaches that host. Keep this separate from a model running entirely on the phone or laptop in your hand.
Which tasks remain online?
| Task | Connection dependency |
|---|---|
| Download a new model | Reach the model source |
| Search current web content | Reach web services |
| Read a new cloud email | Reach the email service |
| Use an online integration | Reach the connected service |
| Access a model outside your local network | Reach the remote host |
| Publish or send the result online | Reach the destination |
Local AI can still prepare part of those workflows. The final connected step needs to be stated honestly.
Are paid features automatically offline?
A paid tier and offline operation are separate questions. Some local features can use downloaded resources after setup, but you must check their platform support, activation needs, and complete workflow.
For example, background capture is opt-in and has its own setup and visible state. It should not be assumed active just because the app is installed. Linux core packages in the v0.0.54-beta.108 release do not establish that every Pro feature is available on Linux.
Use the current release notes for availability instead of assuming parity from a broad feature list.
Check your own offline boundary
After setup, turn off internet access and complete a small real task. Open the local source, load the selected local model, generate an output, check it, and save it somewhere you can reopen without a connection.
If it fails, identify the missing component rather than adding more models at random. It may be a remote selection, an incomplete file, or an online destination.
Download OGAD or get OGAM and prepare one workflow you need. The useful promise is specific: when the input, model, and result stay on a prepared device, supported local work can continue without internet.
Top comments (0)