Cloud Providers
Connect hosted transcription and enhancement services with your own API keys.
Provider Capabilities
Open AI Models → Cloud to connect a provider.
| Provider | Transcription | Enhancement |
|---|---|---|
| Groq | Whisper Large v3 Turbo | Yes |
| Cerebras | No | Yes |
| Gemini | Gemini 3.5 Transcribe | Yes |
| OpenAI | No built-in transcription integration | Yes |
| OpenRouter | Models loaded from its transcription catalog | Yes |
| Anthropic | No | Yes |
| Mistral | Voxtral | Yes |
| Deepgram | Nova 3, Nova 3 Medical | No |
| ElevenLabs | Scribe V2 | No |
| Soniox | Soniox V5 | No |
| Speechmatics | Speechmatics | No |
| AssemblyAI | Universal-3.5 Pro, Universal-2 | No |
| xAI | Grok Voice Transcribe 2.0 | No |
| Cartesia | Ink 2 | No |
These are the integrations and display names in VoiceInk. Your provider account determines which models and usage are available. OpenAI-compatible transcription endpoints can be added through Custom Models.
Connect a Provider
- Select the provider under AI Models → Cloud.
- Enter its API key.
- Click Verify.
- Open a Mode and choose its transcription model or enhancement provider and model.
VoiceInk stores provider API keys in macOS Keychain. Provider panels link to API key setup when available. Removing a key makes workflows using that connection unavailable until you reconnect.
Your VoiceInk license does not include provider charges or credits. Transcription sends audio to the selected service. Enhancement sends text and any enabled context; see Privacy and Data.
OpenRouter
OpenRouter has separate transcription and enhancement model catalogs. Use Refresh Models in the appropriate section to load available models. A model appearing in one catalog does not make it available for the other task.
OpenRouter transcription is batch-based. Its integration does not stream partial text or pass the Vocabulary list as transcription hints.
Enhancement Model IDs
Groq, Cerebras, Gemini, Anthropic, OpenAI, and Mistral support entering a model ID through Custom… in the Mode's enhancement model picker. The ID must be supported by your connected provider; entering it does not create or provision a model.
Real-time and File Support
Streaming support is model-specific. AssemblyAI Universal-2 and Groq's Whisper model are batch-only. Cartesia Ink 2 is streaming-only and requires Real-time for live dictation; it cannot transcribe imported files.
See Real-time Transcription for supported models and Transcribe Audio Files for the file workflow.