Edgee Codex Compressor

The tool inserts a compression layer between a client and the Codex language model, reducing the amount of context that must be sent on each request. In benchmark tests on a code repository, the gateway cut fresh input tokens by roughly half and raised the cache‑hit rate, leading to a 35.6 % reduction in total session cost while preserving output quality.

Developers who run automated code‑analysis or code‑generation workflows with Codex can use the gateway to lower token usage and expense, especially in long‑running sessions where repeated context is re‑sent to the model. The system works with the same model version (gpt‑5.4) and integrates via a simple proxy, requiring no changes to the underlying Codex calls.

What distinguishes this solution is the demonstrated token‑compression efficiency: input tokens drop from 1.14 M to 0.57 M, cached tokens decrease slightly, and cache hits improve from 76 % to 85 %. The result is a cheaper, more frugal Codex deployment without truncating responses.