AI & Models settings
Settings → AI & Models is where you connect cloud provider API keys and install or manage on-device speech runtimes. Individual dictation profiles (and Dictation → Models) choose which connected engine or language model to use.
Cloud providers
API keys are stored locally on this computer (and encrypted if you choose to enable Sync across devices):
- Under Add provider, pick a provider tile.
- Paste your API key, click Test to verify connectivity, and click Save.
- Use Get API key ↗ on open provider cards to jump straight to that vendor's key console.
- Expand any connected card later to Replace, Test, or Remove the key.
| Provider | Speech-to-text | AI polish & Commands |
|---|---|---|
| Groq | Yes | Yes |
| OpenAI | Yes | Yes |
| Google AI Studio | Yes | Yes |
| Deepgram | Yes | — |
| AssemblyAI | Yes | — |
| ElevenLabs | Yes | — |
| xAI | Yes | — |
| Anthropic | — | Yes |
| OpenRouter | — | Yes |
For provider recommendations and step-by-step key setup, see Cloud providers.
Local providers (on-device)
Install an on-device runtime first, then download and start any compatible speech model:
| Runtime | Available models |
|---|---|
| Sherpa-ONNX | SenseVoice Small, Parakeet TDT 0.6B, Moonshine Small |
| Whisper.cpp | Whisper Base |
For each model row you can view download size, supported languages, approximate memory usage while running, hardware compatibility warnings, and Download, Start, Stop, Restart, or Delete model controls.
When Sensitive apps matches a password manager or protected window, Kalam requires a working on-device speech model and refuses to fall back to cloud transcription. If no local model is installed yet, install one here before dictating into protected windows.