Fixing What Providers Broke Underneath Us
Most of v3.4.0 is a direct response to two AI providers changing their model catalogs out from under every app that uses them â including this one. The rest is real feature work: customizable shortcuts, an audio self-test, and more AI providers wired up for real.
đ Gemini: "Invalid API key" with a key you know is correct
Google AI Studio now issues a newer "AQ."-prefixed key format by default, and quietly retired several model IDs for that format â including gemini-2.0-flash, which was this app's default model for every new user, and gemini-2.5-flash, which the Settings "Test Key" button used. A perfectly valid key looked invalid because the model behind it was gone, not the key.
Fixed: Switched to gemini-3.5-flash / gemini-pro-latest / gemini-3.1-flash-lite everywhere, and switched key transmission to the header method Google now documents. Existing users with the old model already saved get migrated automatically on next launch â you don't need to reselect anything.
đ Groq errors â and a shutdown that was 3 days away
Groq's own deprecation schedule retires llama-3.3-70b-versatile and llama-3.1-8b-instant on August 16, 2026 â both were this app's default Groq model and its Test Key model.
Fixed: Removed both, the free openai/gpt-oss-120b and openai/gpt-oss-20b models are now the only Groq options (they already worked fine â this app just wasn't defaulting to them).
đ Screen-share leak â the remaining gap
v3.3.5 fixed the leak that happened on every audio re-enable by caching the screen source after first use. What was left: the very first capture of a session still ran a live remove-stealth â enumerate â reapply dance, which could land while you were already sharing.
Fixed: That first capture is now pre-warmed right after the app launches, well before you're likely to be sharing anything â the live handler is now almost always a cache hit from the start.
⨠Keyboard shortcuts you can actually remap
Every shortcut in Settings is now genuinely customizable â click any combo, press your new keys, done. Conflicts with an already-used combo are caught before they're saved, and a Reset button restores the defaults.
⨠Audio Diagnostics self-test
A new "Run Audio & Deepgram Test" button in Settings walks through exactly what the real interview-audio pipeline does: captures system audio, shows a live level meter so you can confirm it's actually picking up sound, then tests your Deepgram connection separately. If something's wrong, you'll know which half broke instead of guessing.
⨠OpenAI, Anthropic, and Grok are live
These were always fully built â real streaming API integrations â but never connected to the provider picker, so they sat in Settings as "coming soon" while silently collecting keys nobody could use. They're real, selectable providers now, with current model lists (GPT-5.6, Claude Sonnet/Opus 5, Grok 4.6).
đĨ Update Now
All of the above is live in v3.4.0 â completely free, as always, no subscription required.
