Local models and BYOK remote providers

Steps

1. 1.Pick a local ASR path first

   Local options include Qwen3-ASR and other MLX Audio models, Whisper, Cohere Transcribe, and Apple System Dictation. System Dictation is the lowest-friction start if you do not want to download a model yet.

2. 2.Add a local LLM when you want on-device cleanup

   A local LLM can handle punctuation cleanup, concise rewrites, translation, and repeatable prompts offline. Catalog families include Qwen, GLM, Llama, Mistral, and Gemma builds suited to Apple Silicon.

3. 3.Connect remote ASR or LLM with your own keys

   Remote ASR examples include OpenAI Transcribe, Doubao ASR, GLM ASR, and Aliyun Bailian ASR. Remote LLM examples include Claude, Gemini, GPT, Ollama, OpenRouter, Grok, GLM, and Qwen. You pay those providers directly—Voxt does not sell hosted usage today.

4. 4.Keep ASR and LLM independent

   Use a fast or local ASR for capture, then route only rewrite/translation to a stronger LLM when needed. That keeps everyday dictation private and responsive without giving up quality when you ask for a harder transform.

Tips

• Review each provider’s privacy and retention policy before enabling BYOK.
• Store and rotate API keys carefully; treat them like passwords.
• Leave heavy reasoning to remote models while everyday cleanup stays local.
• Need provider-specific setup? See the Remote models doc linked from About.