Your local AI stops answering. Before you delete models or reinstall the app, check whether the component that handles the request is running.
OGAD (Off Grid AI Desktop) includes a System health panel with component status and restart controls where supported. It gives you a focused way to recover a local engine and try a small request again.
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.
Find the part that stopped
A text reply, generated image, and transcription can use different local components. A problem with one does not necessarily mean all of OGAD has failed.
The core health panel in OGAD 0.0.51 is available on Mac and Windows. It reports component state and details. A component that is idle or not installed needs a different response from one that is down.
Start by noting what failed: chat, image generation, spoken output, or transcription. Also check whether you selected a local model or a remote server. Restarting a local engine will not repair an unreachable remote computer.
Recover one component and make a small request
- Stop the unfinished request if the chat still lets you do so. Save any partial text you need.
- Open Settings → Setup & health and find System health.
- Select Refresh to read the current state.
- Find the component for the feature that failed and read its status and detail.
- If its Restart button is available, use it and wait for the result.
- Return to the feature and try a small request with the intended local model.
For chat, use a new short prompt such as “Write one sentence about keeping notes.” For transcription, try a short recording you already have. A small request makes it easier to see whether the component is working again.
The panel only offers restart where the component supports it and its state permits it. Do not expect a restart button beside every item.
Match the fix to the status
| What you find | What to do next |
|---|---|
| A component is starting | Give it time to finish loading before sending more work |
| A component is down and Restart is available | Restart it, then retry one small request |
| A required model is missing | Finish its download and select it before trying again |
| A permission is denied | Review the relevant system permission for that feature |
| A remote server is selected | Check that server and the network connection |
If a small request works but the original one fails, reduce the original workload. A long conversation, large input, or model that leaves little free memory can need more resources than a short check.
Try a smaller compatible model if the selected one cannot run within your computer’s available memory. Close other memory-heavy work before repeating a large request.
Avoid making recovery harder
A component restart can interrupt work that uses it. Wait for another active task to finish, or stop it yourself, before restarting its engine.
The panel also has Free engine, which stops the chat engine and frees its model port where possible. That is a stop action, not a repair button. Use the component’s Restart control for this recovery path.
If the problem remains, keep the exact error text, the model name, and the operation that failed. Those details are more useful for a support report than repeated restarts. Do not delete your chats or models as a first step.
Open OGAD, check System health, and retry one small local request after the relevant component is ready.