Headroom
Compresses what agents read so models use fewer tokens.
- AI tool
- Free
Pricing
Free and open source. License: Apache-2.0.
Overview
Headroom is an open-source layer that shrinks the material an AI agent reads before it reaches the model, including tool output, database results, file reads, logs and retrieved chunks. It is available as a library, a proxy server and an MCP server, with integrations listed for Claude Code, Codex, Grok Build, OpenClaw, LangChain, CrewAI and other SDKs. The project says answers stay the same while token use drops, and it offers reversible compression so the original can be recovered. Its documentation also covers corporate networks and local model setups.
Key features
- Library, proxy server and MCP server
- Compresses tool output, logs, file reads and retrieved chunks
- Reversible compression
- Integrations for Claude Code, Codex, Grok Build and more
Use cases
Alternatives
Last updated 2026-10-20