fix: select newly added model after saving in settings

[tongshu2023]

Summary

Fixes #542 — After configuring a new provider (e.g. DeepSeek) and adding a model, the landing page still shows "Set up model" and prevents entering the classroom.

Root Cause

When a user adds a custom model in settings via handleSaveModel, the setProviderConfig action calls resolveLLMSelection. If the provider has models but no API key yet, it's classified as State A (not usable), leaving both providerId and modelId empty. Subsequent API key configuration correctly resolves the provider but selects the first built-in model instead of the user's custom model. This can cause confusion and a perceived broken state.

Changes

1 file: components/settings/index.tsx (+2 lines of logic, +5 lines of comment)

• Added setModel hook import in SettingsDialog
• In handleSaveModel, after writing models to the store, explicitly call setModel(providerId, modelId) to select the newly added/edited model
• This preserves the user's explicit model choice regardless of credential state

Test Plan

1. Fresh install, open settings, select DeepSeek provider
2. Add a custom model (e.g. deepseek-chat) WITHOUT entering API key first
3. Enter API key, close settings
4. Landing page should show model selector with the custom model selected
5. Enter a requirement and click Enter Classroom — should navigate to generation preview
6. Also verify editing an existing model works correctly

[wyuc]

Thanks @tongshu2023. Closing this.

The reported #542 symptom (the "Please select a model" toast + forced settings reopen) was already fixed by #581 on 2026-05-18. Verified on main: after adding a custom DeepSeek model + API key, Enter Classroom is enabled and navigates to /generation-preview cleanly.

The remaining behavior this PR addresses (picker not auto-switching to the newly-added model) is a deliberate UX choice on our side, not a bug. Adding a model is a configuration action; choosing which model to use is a separate explicit picker action. Auto-switching would also introduce a regression in the common case where another provider is still active and usable, and the user is just adding a model to a side provider for later.

Sorry for the round-trip and thanks for the careful root-cause writeup. #542 is already closed.