Let users choose the writing provider

[wanming]

Requested by Tom · project thread

Before: Inklet only worked with an OpenAI API key. The app already had presets for 20 other providers, but Settings and the saved config always forced writing back to OpenAI.

After: General settings has a Provider picker. The choices are OpenAI, Anthropic, Gemini, DeepSeek, Qwen, Kimi, GLM, and the other existing presets, plus Custom (OpenAI Compatible) with an Endpoint field. Writing and selection translation use the chosen provider and its own key. Realtime dictation and pronunciation still use OpenAI, so when another provider is selected, a separate optional OpenAI API Key field appears for those features. With OpenAI selected, the screen works exactly as before: one key covers everything.

This removes the biggest setup blocker for people who can't easily pay for OpenAI, which is the first step in the growth plan for the project.

How

• AppConfig now keeps the saved providerID. Configs saved before version 5 still resolve to OpenAI, so stale provider IDs from early builds don't switch anyone's provider when they upgrade. Unknown IDs also fall back to OpenAI. resolvedProviderPreset applies the custom endpoint when it's a valid http(s) URL.
• SettingsViewModel loads and saves each provider's key under its own Keychain account. Switching providers first flushes any key you were typing, then loads the new provider's key and default model. The model list now comes from the selected provider's catalog.
• Onboarding checks for the selected provider's key, not the OpenAI key.
• The UI reuses the existing Provider, API Key, and Endpoint strings, and their help text is rewritten. Two new keys are added in all ten languages.
• README (English and Chinese) and the manual QA checklist are updated.
• I didn't touch the popover or dictation code. One known gap: the popover header shows the custom preset's English name.

Verification

• I added config decoding and round-trip tests, plus Settings view model tests for switching providers and saving keys per provider. I couldn't run swift test here (no Swift toolchain), so CI is the first run.
• Manual QA still needed: switch to another provider and transform and translate with it, check that dictation and pronunciation still use the OpenAI key, try a local custom endpoint, switch back to OpenAI, and check that the General panel fits in English and Chinese.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KrJfbzXfbZ33kLGiZbfpz7

---

Generated by Claude Code