feat: add optional cloud transcription via Groq (free Whisper Large v3)

[nicremo]

Summary

Adds cloud-based speech-to-text as an optional alternative to the local Whisper pipeline. The existing fully-local mode continues to work unchanged.

• Cloud transcription via Groq: Uses whisper-large-v3 for significantly better accuracy than local whisper-base
• Completely free: Groq offers a generous free tier (7,200 audio seconds per hour)
• Automatic offline fallback: When internet is unavailable, falls back to local Whisper seamlessly
• 3 modes: Auto (cloud + fallback), Cloud-only, Local-only (original behavior)
• Any OpenAI-compatible provider: Works with Groq, OpenAI, Lemonfox, or any compatible API
• API key security: Encrypted via macOS Keychain (Electron safeStorage), never stored in plaintext

Why Groq?

Provider	Price/min	Model	Free Tier
Groq	$0.0002	Whisper Large v3	7,200 sec/hr
OpenAI	$0.006	Whisper v2	None

Groq is 30x cheaper than OpenAI and the free tier covers normal dictation usage entirely.

What changed

• New files: api-key.ts (encrypted key storage), cloud-transcription.ts (OpenAI-compatible API client)
• Modified: dictation.ts (transcription routing), App.tsx (TranscriptionCard UI, updated setup wizard), types.ts (new settings fields)
• No breaking changes: Default mode is "Auto", which tries cloud first and falls back to local. Users without an API key get the exact same local-only experience as before.
• No new npm dependencies: Uses native fetch, FormData, and Blob (Node 22+ in Electron 41)

UI additions

• Transcription card on the Models page: source selector (Auto/Cloud/Local), API key input, cloud model dropdown, language selector
• Updated setup wizard: New "Transcription Engine" step for choosing cloud vs local
• Relaxed validation: Ollama is no longer required when enhancement level is set to "No Filter"

Test plan

• Fresh install: setup wizard shows transcription engine choice
• Enter Groq API key, verify it saves and encrypts (check settings.json for openaiApiKeyEncrypted)
• Cloud mode: Fn dictation transcribes via Groq, text is polished by Ollama
• Disconnect internet: Auto mode falls back to local Whisper
• Local mode: behaves exactly like the original app
• No API key: app works in local-only mode without errors

[cekimilf]

Amazing! But can a version of this fork be made for windows as well?

[nicremo]

@cekimilf I will try to cook something together. Check out https://github.com/nicremo/openwhisp-enhanced

[cekimilf]

@cekimilf I will try to cook something together. Check out https://github.com/nicremo/openwhisp-enhanced

Amazing, thank you! I will check it out once it's ready! Already downloaded your version, looks really good. I like how you can play around with settings and choose whichever model and approach you prefer, and how you can change your keybinding. (I like the Ctrl + Shift + Space like in the default program)

Would love to use this for both English, Serbian (and German, which I am slowly learning)

[nicremo]

@cekimilf can you hit me up via Mail? kontakt@bitzer-fabian.de
Created a Windows Version but need someone to test in on windows.