AI Assistant Settings
The AI assistant settings page configures the connection and models used by the chat service. Open it from the AI hub via the AI Assistant settings entry. API keys are stored in the system’s secure storage only; they are never written to ordinary settings files.
The page is organized into the following sections.
Connection
- Provider — choose OpenAI-compatible, Anthropic, or DeepSeek. Each provider keeps its own endpoint, model, and model list. Note that tool features such as manual search require a model with tool-calling support (e.g.
deepseek-chat;deepseek-reasonerand some self-hosted servers do not support it). - Endpoint — the base URL of the service; leave empty to use the provider’s default.
- API Key — enter your API key. Once configured the row shows “Configured” (only that, never the key itself), with a clear-key button.
Models
- Default model — pick the model used by the current provider from the saved models.
- Fetch models from server — asks the provider for its model list and saves them as choices (the first model is auto-selected if none is chosen yet). Fetches share a 30-second cooldown, so the action is temporarily disabled after a request; a missing API key is reported. The number of fetched models is shown in a snackbar.
- Model list — lists the saved models (with owner info when known) with a remove button each, plus Add model to type a name manually.
Actions
- Test connection — sends a test message to the current configuration to verify endpoint, key, and model.
- Clear conversations — deletes all saved conversations after confirmation.
- Re-download AI data — force-refreshes and re-downloads the assistant’s resource database, useful when the data is corrupt or behaving oddly.
Disable the AI Assistant — confirms and turns off the global assistant switch, returning the gate to the “Enable” state; re-enabling does not require the notice to be read again.
The AI assistant is available on native platforms (Android, desktop) only; the web app does not include it. For how to converse and what the assistant can do, see the chat page.