It would be great to have visibility into token usage during a Goose session, specifically:

1. Track and display the number of tokens used in the current session
2. Implement warning system for high token usage that might lead to problems
3. Potentially show:
   • Current token count for the session
   • Token usage per interaction
   • Warnings when approaching model context limits
   • Estimated cost (if applicable)

This would help users:

• Monitor resource usage
• Avoid unexpected cutoffs due to context limits
• Better manage costs when using paid models
• Optimize their interactions for token efficiency

The warnings could be configurable based on the model being used, since different models have different context limits.