AI Models
Choose the models for dictation and AI Processing, and add your own API keys.
The AI Models screen sets two things:
- Dictation: the model that turns your voice into text.
- AI Processing: the AI that edits the text in modes with AI Instructions. This part is optional.
Open AI Models
- Mac, Windows and Linux: click AI Models in the sidebar. In Modes, the AI Models button opens the AI Processing part.
- iPhone and Android: tap the model name on the Transcribe screen. In Modes, the key button opens the AI Processing part.
Model Types
Each part has tabs:
- Spokenly Cloud: models on Spokenly's servers, included with Spokenly Pro. With Local+, this tab is not shown.
- Local Models: models that run on your device. For AI Processing, this tab has Ollama and LM Studio on Mac, Windows and Linux, and Apple Intelligence on supported Macs and iPhones. On iPhone and Android, Ollama and LM Studio are in the provider list of Your API Keys.
- Your API Keys: models from your own account with a provider. The provider bills you directly, and you do not need Pro.
Add an API Key
- Open Your API Keys.
- Pick a provider in Add Provider. When you have no keys yet, the tab also shows Groq, OpenAI and ElevenLabs for dictation, and OpenAI, Google Gemini and Groq for AI Processing.
- Paste the API key. Get API key opens the page of the provider where you create a key.
- Type a model name or pick one in Browse. When the provider publishes its model list, Browse shows the full list. Otherwise it shows a few tested models. For dictation, an empty Model field uses the default model of the provider.
- Select Test & Save. Spokenly sends a short test request. When the test passes, Spokenly saves the key and selects it.
If the test fails, Spokenly shows the error and does not save the key.
Providers
Dictation: Groq, OpenAI, ElevenLabs, OpenRouter, Google Gemini, Deepgram, Mistral AI, AssemblyAI, Cartesia, Cohere, Fireworks AI, Sarvam AI, Straico, xAI, and any server with an OpenAI-compatible API. Mac and iPhone also have Corti and Volcengine.
AI Processing: OpenAI, Google Gemini, Groq, OpenRouter, Anthropic, Cerebras, Inception, Mistral AI, Straico, xAI, any server with an OpenAI-compatible API, Ollama and LM Studio. The Mac also has Amazon Bedrock.
One Key for Both Parts
OpenAI, Groq, Google Gemini, Mistral AI, xAI, OpenRouter and Straico keys work for dictation and for AI Processing. When you add one of these providers and the other part already has a working key for it, Spokenly fills in the key for you.
Several Keys for One Provider
You can add the same provider more than once, for example a work key and a personal key. When two keys use the same model, the list shows the last four characters of each key.
Edit or Delete a Key
On iPhone and Android, select the key and tap Edit under it. On Mac, Windows and Linux, use the menu at the end of the row.
If the last test failed, the key shows Connection failed. Edit the key and test it again.
Delete removes the key. When you delete the dictation key that is in use, Spokenly uses the default dictation model.
Models in Modes
Each mode can have its own Dictation Model and Text Model in its settings. A mode without its own model uses the models selected on the AI Models screen. In Select Text Model, pick the selected model again to make the mode use the AI Processing model of the AI Models screen. A mode with its own text model can also have a Fallback Text Model, which Spokenly uses when the main model is not available.
Sync
With Sync Settings and the API Keys category on, your keys sync to your other devices that are signed in to the same account.