A model needs a compatible runtime
A model name is not enough to establish compatibility. The file format, architecture and quantisation must be supported by the application’s runtime. Start with the models offered by your installed application or its current compatibility list.
Local processing means the model runs on the device. It does not guarantee that every optional tool, speech service or search request stays offline.
Leave room for the operating system
Model file size and memory use are not the same. Inference also needs runtime memory and space for the conversation context. Other applications compete for the same resources.
If loading fails or the system becomes unresponsive, use a smaller supported model and a shorter context. Compare models with the same short task on your device rather than inferring speed from parameter counts alone.
Make the first test simple
- Download the compatible model completely.
- Confirm the app has loaded it.
- Ask a short fresh question.
- Check whether the answer is useful and whether the device remains responsive.
- If offline use matters, repeat with the network disabled.
For an error report, include the model name, application version, device and available memory. Keep credentials and private conversation content out of public reports.
Local AI in Denny
Denny includes a local model setup flow. You can also choose supported external AI providers using your own account and API key, or use a supported local connection such as Ollama. Availability depends on the installed build.
The model catalogue (RU) lists examples, sizes and compatibility notes. Current full Denny for Mac requires macOS 14 and Apple Silicon; confirm Android access and the appropriate model with support.
