Skip to main content

AI & Models settings

Settings → AI & Models is where you connect cloud speech and language providers (API keys) and download on-device speech engines. Profiles on the Dictation page choose which provider and model to use; this tab stores credentials and local engines.

What you see

Two sections:

  1. Providers & API Keys — connect Groq, OpenAI, Anthropic, Google Gemini, OpenRouter, and optional custom endpoints
  2. Local engines — download, start, stop, or delete SenseVoice Small, Whisper Base, Parakeet TDT 0.6B, and Moonshine Small

Providers & API Keys

If nothing is connected yet: No providers connected yet. Use the dropdown below to add one.

Add or manage a provider

  1. Under Add provider, choose a service and click Add.
  2. Paste the API key, optionally Test, then Save.
  3. Connected providers show a masked key with Replace key and Remove.
ProviderTypical use
Kalam CloudHosted speech and AI on Max (authenticated with your Max license — no separate API key).
GroqFast cloud STT and LLM (bring your own key).
OpenAICloud STT and LLM.
AnthropicCloud LLM (for polish / language steps).
Google GeminiCloud LLM.
OpenRouterCloud LLM; pick models from OpenRouter’s catalog in the profile editor.

Each provider row notes whether it supports speech (STT), language models (LLM), or both. Many providers include a Get API key ↗ link while editing.

Custom OpenAI-compatible endpoint

SettingWhat it does
Custom OpenAI-compatible endpointOptional base URL and API key for profiles that use the Custom endpoint provider. Pick the model on each dictation profile.
Base URLOpenAI-compatible root (for example https://api.example.com/v1).
API keyKey for that endpoint.
Where models are chosen

API keys live here. Which provider and model a dictation uses is set per profile on the Dictation page (voice provider and language model). See Modes and AI polish.

Local engines

Download and manage on-device speech engines here. Choose which engine a profile uses on Dictation (voice provider → Local).

SettingWhat it does
Installed enginesList of local STT models with size, quality, languages, and status

Local engines

For each engine you can:

  1. Download (when not installed)
  2. Start / Stop / Restart when installed
  3. Delete to remove the model files
EngineBest for
SenseVoice SmallDefault local; multilingual
Whisper BaseBroad languages (whisper.cpp)
Parakeet TDT 0.6BHigh-accuracy English (CC-BY-4.0)
Moonshine SmallFastest local English

Status badges include states such as Running, Stopped, Starting, and Error. When running, approximate RAM use may appear. Hardware warnings appear if your machine cannot run a model, or if the engine is not available on this platform.

If a local speech engine runtime is installed, you may also see Uninstall engine.

Sensitive apps need a local engine

With Sensitive app detection on, matching apps force on-device STT. If no local engine is installed and running, dictation fails instead of sending audio to the cloud. See Privacy settings and Local models.

Plan gates

FeatureFree / ProMax
BYOK providers (Groq, OpenAI, …)Yes (your keys)Yes
Custom OpenAI-compatible endpointYesYes
Local enginesYesYes
Kalam Cloud hosted STT / AINoYes (included in Max)