Headroom

About

Pay for the tokens your agent actually needs

Run headroom wrap claude (or codex, cursor, grok, opencode, aider and more) and every session routes through a local proxy that compresses what the agent reads. The README reports about 20% fewer tokens for coding agents and 60 to 95% fewer on JSON heavy tool output, with the same answers on its benchmarks. Compression is reversible: originals stay in a local cache, and the MCP server exposes headroom_compress, headroom_retrieve and headroom_stats. Extras include a shared memory store across agents and headroom learn, which reads failed sessions and writes corrections into CLAUDE.md, AGENTS.md or GROK.md. Things to know: an anonymous usage beacon (compression ratios, model IDs, OS, never prompts or code) is on by default and turns off with HEADROOM_TELEMETRY=off or DO_NOT_TRACK=1, and headroom wrap also installs Serena for code navigation unless you pass --code-memory none. Created by Tejas Chopra, Apache 2.0, about 76k GitHub stars and trending on Oct 19, 2026.

Features

• Compression runs locally, nothing is sent out to be compressed
• One command wrap for Claude Code, Codex, Cursor, Grok and more
• Drop-in proxy with no code changes
• MCP server with compress, retrieve and stats tools
• Reversible: originals cached and fetched on demand
• headroom learn turns failed sessions into AGENTS.md fixes

Install