Troubleshooting
Models
Section titled “Models”“This file stores weights in a format (tensor type N) that the local runtime cannot read” : The file was made for a modified llama.cpp (for example Prism ML’s PQ2_0). Choose another quantization of the model. Dismiss clears the error.
“The local runtime does not support the X model architecture yet” : The model is newer than the bundled llama.cpp. A later release will support it.
“Not enough GPU or system memory to load this model” : Unload other models, lower the context size in the model settings, or choose a smaller quantization.
A model loads slowly or answers slowly : Check the strip at the top of Models. If part of the model is in RAM, a smaller quantization that fits the GPU is much faster.
“This local model has no vision adapter”
: Images need a vision model. Download a variant with an mmproj file, or attach one with Add vision adapter.
Windows uses the integrated GPU or the CPU : Integrated graphics are skipped on purpose. On an NVIDIA PC, install CUDA from Settings → Local Runtime; if it can’t use the GPU, the card shows why (usually an old driver).
The server runs models on the CPU
: sudo local-cognitive-server status says why. See NVIDIA GPUs.
A new GPU in the server is not used
: Reboot after installing it, then check nvidia-smi and Models → GPUs.
Remote
Section titled “Remote”The server shows as offline
: Run sudo local-cognitive-server status on the server. It needs outbound HTTPS to api.local-cognitive.com; logs shows connection errors.
“Update Local Cognitive on the server to chat”
: The server is older than the app. Run sudo local-cognitive-server update.
The connection key doesn’t work
: Keys expire after 10 minutes and work once. Run pair again. Check that you’re signed in to the account that owns the server.
Updates
Section titled “Updates”The Mac says it can’t install the update : Move the app to Applications and start it from there; updates can’t replace an app running from the dmg.
Windows doesn’t install updates itself : Expected for now: download the new installer from the release page. See Updates.
Reporting a problem
Section titled “Reporting a problem”Use Report a bug (the bug button in the title bar, or Settings → Report a bug). With error reports turned on in Settings → Data & Privacy, the report carries the diagnostics that help find the cause. You can also open an issue on GitHub.