bug: Token Count shows 0% when model is not loaded

[louis-jan]

Problem

The token count displays 0% when the model is not loaded. This provides no useful information to users and creates a poor user experience, especially when users are composing messages in a thread.

Steps to Reproduce

1. Open a thread without loading the model
2. Start typing a message
3. Observe the token count indicator shows 0%

Expected Behavior

One of the following should occur:

• Model should be automatically loaded when a thread is opened, OR
• Token count should show an estimated value based on a default model's context size, OR
• Display a clear indicator that the model needs to be loaded for accurate token counting

Actual Behavior

Token count shows 0%, providing no useful feedback to users about their message length relative to context limits.

Proposed Solution

Implement automatic model loading on thread load. This would:

• Provide immediate, accurate token count feedback
• Improve user experience by eliminating manual model loading step
• Ensure users are aware of context limits before composing long messages

Impact

• Users cannot gauge message length against context limits
• Reduced usability when composing messages
• Requires manual model loading for basic functionality

🤖 Generated with Claude Code

Co-Authored-By: Claude noreply@anthropic.com

[louis-jan]

Another path is to wait for FFI refactoring @qnixsynapse