Default Model Selection

[marksftw]

The default LLM selected for the user can vary depending on the scenario. Here are a few situations to handle:

1. New user on the Free plan. Their only model is Llama 3.3
2. New user on a Starter plan. They have access to Gemma 3. Select that one as default.
3. New user on Pro, Max, or Team. They have access to all of the models. Select GPT-OSS 120B as the default.
4. Existing user who has already chatted, select the last model used. Be sure to persist this in the user's profile values so it can be recalled on next launch of the app.

[marksftw]

@claude build this

[claude]

Claude finished @marksftw's task —— View job • claude/issue-184-20250809-1954 • Create PR ➔

---

Default Model Selection Implementation Complete! ✅

I've successfully implemented the default model selection feature based on user plans and usage history.

Todo List:

• Explore codebase structure for user management and model selection
• Identify current model selection implementation
• Implement default model selection for different plan types
• Add persistence for last used model in user profile
• Test the implementation
• Run linting and build commands
• Commit and push changes

Implementation Summary:

• Free plan users: Default to Llama 3.3 70B
• Starter plan users: Default to Gemma 3 27B
• Pro/Max/Team plan users: Default to GPT-OSS 120B
• Existing users: Last used model takes priority over plan defaults
• Persistence: Last used model is saved to localStorage via OpenSecret storage

Changes Made:

1. Created /utils/modelDefaults.ts with plan-based default model logic
2. Updated LocalStateContext to load/save last used model and apply plan-based defaults
3. Simplified model access checking in ModelSelector to use shared utility
4. Modified setModel to persist the selection automatically

Technical Details:

• The system checks for existing usage history first, then falls back to plan-appropriate defaults
• Model selection is automatically persisted when users manually change models
• Plan detection works with the existing billing system integration
• All code passes linting and builds successfully

---