headroomlabs-ai/headroom
59,530 4,419PythonApache-2.0Updated 3 months ago
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
Our verdict: worth keeping an eye on
Verified real 2026-08-23: 24k weekly npm downloads, Apache-2.0, 5.2k forks. Proxy (headroom wrap claude) or MCP; compresses tool output before it enters context. Shrinks bytes-per-turn, NOT turns - measured root cause on this box is step count x full-context resend (181 responses, 48.8M cache reads), so expect ~10-20% off, not the 84%. Watch until the step-count levers (delegation, /clear at 150k) are actually used.
Filed under context tooling in our directory.