Headroom

About

Headroom compresses everything your AI agent reads — tool outputs, logs, RAG chunks, files, conversation history — before it reaches the LLM, with measured accuracy preserved on benchmarks like GSM8K and TruthfulQA. Works as a Python/TypeScript library, a drop-in proxy, an agent wrapper (Claude Code, Codex, Cursor, Copilot, Cline, and 15+ others), or an MCP server. Includes cross-agent shared memory, reversible compression (CCR), output token reduction, and a `headroom learn` failure-mining feature. Local-first: your data never leaves your machine.

Topics