TokenTamer A proxy that reduces LLM token usage through context compression

I built TokenTamer, an open-source proxy that sits between AI coding assistants and LLM APIs.

The goal is to reduce token consumption before requests reach the model by applying techniques such as:

Context deduplication
Conversation compression
Intelligent summarization
Smart context filtering

I originally built it after noticing that coding agents often resend large amounts of repeated context, leading to unnecessary token usage and higher costs.

TokenTamer is designed to be lightweight and easy to place in front of existing workflows.

I'd love feedback on the architecture, compression strategies, and potential use cases.(https://github.com/borhen68/TokenTamer)