Current Implementation: repo-contextr currently uses a simple character-based approximation (~4 characters per token) for token counting. While this provides quick estimates that work reasonably well for English code and documentation, it's not perfectly accurate for real-world LLM usage.

Future Enhancement: We plan to integrate OpenAI's tiktoken library to provide accurate token counts that exactly match GPT-3.5, GPT-4, and other OpenAI models. This will give you precise token estimates for context planning, cost estimation, and optimization decisions when working with LLMs.

This issue was inspired by the implementation of the token count feature in repomix.