Show HN: Token compression CLI to save Codex/Astra costs

Category: ai-ml

Tags: token-compression, llm-costs, codex, proxy, ai-agents

Score: 5.3/10 (Innovation: 7, Technical: 5, Documentation: 4, Utility: 5)

A CLI tool that compresses Codex tool output before it enters the model context by routing through a hosted service running a fine-tuned Qwen model. It claims up to 30% token reduction while preserving agent trajectory and avoiding KV cache invalidation through a proxy architecture. The core compression model and service are proprietary, so the repo mainly contains the client-side integration and installer.

Target audience: AI developers and engineers using Codex who want to reduce LLM token costs

Repository: https://github.com/spenmcke/compress · Python · MIT · 67 stars

View on Hacker News