Headroom
Local compression layer that shrinks LLM context before it is sent
What it is
Headroom reduces the token count of tool outputs, logs, files, RAG chunks and chat history before they reach a language model, using separate compressors for JSON, source code and prose. It runs on your machine as a Python or TypeScript library, an OpenAI/Anthropic-compatible proxy, an MCP server, or a one-command wrapper around coding agents such as Claude Code, Codex and Cursor. Originals are cached locally so the model can retrieve full text when needed. Developers running coding agents or LLM apps use it to cut API costs and keep long sessions inside the context window.