You want local AI, but you do not want your first evening to become a comparison of model names and file formats.
OGAD (Off Grid AI Desktop) can suggest a model setup based on your computer’s memory and the amount of hardware you want it to use. You review the plan, select Configure, and start with chat while the remaining model setup continues.
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.
Get a useful starting set of models
Local AI uses model files stored on your computer. Chat, voice, transcription, and image generation can need different files. Guided setup brings those choices into one plan instead of making you select each one from a catalog.
In OGAD 0.0.51, the setup preview shows the planned models, their capabilities, whether they are installed, and the approximate remaining download size. It is a starting configuration, not a guarantee that every feature will run well on every computer.
This setup is available without Pro on Mac and Windows. OGAD beta108 also includes a Windows local speech runtime for English (US and UK) voices. Guided setup includes a speech model, but wait for its resources to finish downloading and prepare your selected voice before playback. The Windows path does not provide the full multilingual voice set available on Mac. Pro features still require their own access and permissions.
Choose how much of your computer to use
The Resource use options let you choose a starting point:
| Option | When it is useful |
|---|---|
| Conservative | You want a lighter setup and more memory left for other work; the plan skips an image model |
| Balanced | You want the recommended starting balance for your computer |
| Extreme | You want a larger model and context allocation and are prepared to give AI more memory |
Start with Balanced if you do not have a specific reason to choose another mode. Use Conservative if you want to begin with a lighter chat workflow while other apps remain open.
Review the actual preview on your computer. The models and total download can vary with available hardware and what you have already installed. Download size is disk space; the memory needed while running a model is a separate requirement.
Get your first local answer
- Install OGAD and open the first-run setup, or go to Settings → Setup & health.
- Find Configure it for me and choose your resource-use mode.
- Read the Will set up list and check the approximate download size.
- Select Configure and keep an internet connection while required files download.
- When chat is ready, open a chat with the installed local model.
- Ask: “Give me a short checklist for planning a weekend walk. Use five points.”
Chat is prepared first. The other model downloads are attempted afterward. Wait for each supported capability’s setup to finish before trying it. A failed optional download does not mean the working chat model failed.
Your first goal is a complete local reply. You can explore other model choices later, once you know which tasks you want to use AI for.
What works without internet afterward?
A downloaded, working local model can process supported requests on your computer without cloud inference. That applies to the local task you configured—not web searches, remote model servers, or connected online services.
If a feature still needs a model download, it is not ready for offline use. Before traveling, finish the downloads and check the local capability you plan to use.
If setup cannot finish
Check the reported setup error, free disk space, and your connection. A large download needs room for its files. A model also needs enough working memory when it starts.
If the chosen setup is too heavy for the way you use the computer, return to the resource-use selection and review a lighter plan. Do not keep adding large models merely because they are available.
Turn setup into one useful piece of work
After the first small reply, try a task you would otherwise send to an online chatbot. Paste a short project update and ask for a clearer version under 120 words, preserving every date and open question. Check the result beside your input.
If that works with the guided model choice, you already have a useful reason to keep the app. You do not need to install several alternatives before doing real work. Try another model only when the current one misses a capability or does not produce acceptable answers.
For a second capability, check the plan and the completed downloads first. If you chose Conservative, it intentionally skipped an image model. If you want images later, select a compatible image model and complete that setup separately. A missing image capability is not evidence that chat setup failed.
Check the setup before leaving the network
Run one small request for each local capability you actually intend to use. Then turn off internet access and repeat with the same local selections. Use questions about supplied text or an idea, not current weather or a web page.
If chat works offline but another feature asks for files, complete that feature’s download while connected. If the downloaded model cannot load, review memory and try a lighter choice. An installed package and a useful running model are different results.
Keep the working choice as your starting point. Guided setup is most valuable when it gets you to a real local task quickly; larger downloads are optional steps for a need you can name.
Try OGAD’s guided setup and begin with one local chat. You can choose more specialized models after that first useful result.