Headroom

SummaryA local library and proxy that shrinks logs and JSON tool output before a coding agent sends them to the model.

Filed October 18, 2026. Price, as published by the maker: Free, open source (Apache-2.0). No account. Paid Enterprise plan for teams..

Headroom 0.41.0 runs on your machine as a Python library, a proxy, an agent wrapper or an MCP server. The library needs no account and no API key. The proxy and headroom wrap pass your own model provider key through. An anonymous usage beacon is on by default; HEADROOM_BEACON=off or DO_NOT_TRACK=1 turns it off.

MODEL ran the library on Linux with no model attached. A 10,145-token log came back at 1,759 tokens with its one error line intact. A 37,700-token JSON response came back at 12,836 tokens with every row kept. A source file came back unchanged, by design. The full test, the catch, and what we did not test are in the story. The savings and accuracy tables in the README are the maker's.