TokenTamer, a middleware proxy for AI coding agents, compresses code context in real-time to cut LLM API costs by 50–80%

TokenTamer is a drop-in proxy that sits between a coding agent and an LLM API, intercepting payloads and replacing background files with structural skeletons using AST parsing while keeping active files and signatures intact.

The tool supports multiple coding agents including Aider, Cursor, Claude Code, and Codex CLI via HTTPS interception or direct API base URL configuration, with tool-aware compression that skeletonizes older file reads while preserving the most recent read of each file.

The project is currently alpha software in active development; savings estimates of 50–80% come from unit tests with synthetic payloads rather than production benchmarks, and compression benefits depend on agents re-reading files multiple times.

Installation requires Python 3.9 or newer and a one-line setup via git clone; for hardcoded endpoints like Claude Code, users must manually trust the local certificate and add API domain entries to /etc/hosts.