Headroom is an open-source context-compression layer for AI agents that reduces the number of tokens sent to and returned from language models. It can compress tool outputs, logs, RAG chunks, files, and conversation history before they enter the model while retaining access to the original content when needed. The project provides a library, drop-in proxy, MCP server, and wrappers for popular coding agents, making it possible to add compression without redesigning an existing agent stack.

Headroom uses content-aware compressors for formats such as JSON, code, and prose, and can maintain shared cross-agent memory with deduplication. It also includes tooling for learning from failed sessions and reducing unnecessary output verbosity. The main benefit is lower token usage and potentially lower latency or model cost, especially for agents that repeatedly consume large structured outputs. It is best suited to developers operating long-running or tool-heavy agents where context size becomes a practical bottleneck.