You want offline AI, without filling your phone first.
The storage you need depends on the models you download and the files you keep. OGAM (Off Grid AI Mobile) lets you start with one local model, then add image or speech models when you need them. Check the download total before you begin. Storage capacity and available RAM are separate limits.
Get OGAM for your phone | Mobile releases
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 takes up space in an offline AI app?
A model is a file, or a group of files, that the app downloads before it can perform a local task. Your chats and output files are stored separately. You can therefore have several models taking up storage even when only one is running.
Think about the complete setup rather than one attractive number on a model card.
| Storage item | Why it matters |
|---|---|
| The app | It must fit before you download models |
| Text model files | Different model sizes and variants have different download totals |
| Vision companion files | Some models need an additional file to process an image |
| Transcription model | Voice input needs speech recognition resources |
| Speech output resources | Reading replies aloud needs a supported voice setup |
| Image or video model files | These can include several components |
| Saved media and documents | Your own inputs and generated output also accumulate |
Downloading every capability at once makes the first setup harder to manage. Choose the task you want most and prepare its files first.
How do you estimate the space before downloading?
Open the model catalog and inspect the size for the actual model you selected. Use the total download when several files belong to the model. Do not estimate from the model's name or parameter count alone.
As a planning example, suppose your chosen text model needs 1.3 GB, its companion resources need 700 MB, and a transcription model needs 150 MB. Those downloads total about 2.15 GB. This is arithmetic for a sample setup, not a claim that every phone receives that bundle. Add space for the installed app, your documents, generated files and download working space.
If another model is 3 GB, keeping both means keeping both sets of files. Switching which model is active does not automatically remove the first download.
There is no single storage figure that fits every offline AI workflow. A short text-writing setup and a collection of image models are different plans. The catalog and the phone's own storage screen give you the relevant numbers for your choice.
Where can you check storage in OGAM?
Open Settings > Storage to review downloaded material and available space. The storage screen includes model categories and download cleanup controls. The phone's system settings provide a second view of total app storage.
Use both when the numbers appear different. A model list describes its model files; the operating system can include additional app data. Do not assume every byte belongs to a model you can see in one category.
The mobile implementation keeps separate records for text, image, video, transcription and speech downloads. That is why a successful chat download does not mean voice or image generation is also ready. OGAM source and releases.
What should you download first?
Start with the smallest setup that can complete a useful task. For example, if you want help writing client updates, begin with one text model. You can test it using a note you already understand:
The first draft is complete. Two photos need replacing. The client will choose them tomorrow. We can send the revised version after that decision.
Ask:
Write a short status update. Separate completed work, the open decision and the next step. Do not add a date or promise that is not in the note.
Review the answer. Then try a second note with different details. If the result is useful, keep that model as your starting setup. You have established a reason to use the storage before adding more downloads.
Next, add transcription if speaking your notes would help. Add an image model when you have an image task. This sequence makes each extra download a deliberate choice.
Does enough storage mean the model will run?
No. Storage holds the downloaded files. RAM holds the working model and the data it processes. A phone can have plenty of free storage but insufficient available memory for a large model.
A 2 GB download does not mean the app will need exactly 2 GB of RAM. Runtime data, input length, image processing and other applications can change the working memory demand. Use the model guidance in your installed app, then test a short request.
If a model downloads but fails to load, deleting unrelated photos may not solve the problem. Try a smaller compatible model, close other demanding apps and reduce the size of the task. If it cannot download at all, storage and connectivity are the first checks.
How can you reclaim space without losing useful work?
Review which models you actually use. Keep a working text model before deleting alternatives. Use the model's removal control for downloads you no longer need, and read the confirmation carefully.
Unloading a model releases working memory; deleting its downloaded files reclaims storage. Those actions solve different problems. Clearing a stale download entry also does not mean you have removed every saved file from the app.
Before a trip, remove optional models only after testing the remaining setup. A deleted model must be downloaded or transferred again before it can run locally. You do not want to discover that missing file after you lose connectivity.
How do you know the phone is ready to work offline?
Complete the downloads while connected, load the chosen model, and get one successful answer. Prepare voice resources too if you intend to speak or listen. Then disconnect and repeat a short task.
If the app requests another download, reconnect and finish that component. Remote models and web searches still need network access; a local model does not make those operations offline.
For a useful final check, use the exact workflow you plan to need: one dictated note, one local reply, or one generated image. Keep enough free storage for the files that workflow will produce.
Open OGAM, choose one useful local model, and check its full download size. You can expand the setup once that first task works.
Top comments (0)