GitHub – chopratejas/headroom: Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 60-95% fewer tokens, same answers. Library, proxy, MCP server.

Headroom is one of those projects that feels boring in the best possible way. It looks at a very practical problem, AI agents are still paying for a lot of text they don’t really need.

You can explore it here: https://github.com/chopratejas/headroom

The idea is simple enough. Headroom compresses tool outputs, logs, files, RAG chunks, and conversation history before they reach the LLM. In plain terms, it trims the fat before the model sees it. According to the project, that can mean 60% to 95% fewer tokens, while keeping the same answers. That’s not a small cleanup. That’s the difference between an agent that feels lean and one that starts dragging its feet as context grows.

What makes this interesting is the shape of the solution. Headroom isn’t just a library, it also ships as a proxy and an MCP server, so it can sit in the path of existing workflows without forcing you to rebuild everything from scratch. That matters. A lot of token-saving ideas sound nice until you ask the team to rewrite half their stack just to try them.

There’s also a systems-minded touch here. Headroom is local-first, reversible, and built around a stable request lifecycle, which means it’s aiming at real operational use, not just a demo that looks clever in a notebook. It supports common agent setups like OpenAI-compatible clients, MCP-native workflows, and integrations with frameworks such as LangChain and Agno. It even includes tools for learning from failed sessions, which is a nice reminder that efficiency isn’t only about shrinking inputs, it’s also about improving the loop over time.

I also liked the restraint in the design. Compression disabled? Fine, it can still act as a gateway. Need it for logs, memory, or code? There’s room for that too. That kind of flexibility is what usually separates a useful infra tool from a short-lived experiment.

If your agents are getting expensive, sluggish, or just annoyingly verbose, Headroom is worth a serious look. It points toward a future where context is still rich, but not wasteful. And that’s a pretty sensible direction.