headroomlabs-ai/headroom
A local layer that shrinks what goes into an LLM. Tool output, logs, RAG chunks and files are compressed by content type, and the original is kept so the model can fetch the full text back when it needs it. It comes as a library, a proxy, an agent wrapper or an MCP server.
RiskyKey files read, not run
README-pypi.md says the proxy and MCP server send an anonymous usage beacon by default (a deploy/beacon collector is in the tree), so usage data goes to a server the user did not choose unless they set HEADROOM_BEACON=off or HEADROOM_OFFLINE=1.
How we know
We read the README and key files, not every source file, and did not run it, so the most we can say is Idea adapted or Risky.
Good for
Anyone whose agent sessions burn tokens on large tool output or repeated context and wants a cut without rewriting prompts.
Not for
Short chats with little context; how much it saves depends heavily on the kind of content you feed it.
Repo
- Stars
- 74,841
- License
- Apache-2.0Use it, change it, sell it, closed-source included. Keep the author's copyright notice.
- Maturity
- active
- Last push
- 2026-10-09
- Last checked
- 2026-11-14