headroom

What it's for.

• Shrink agent tool output and logs before LLM calls
• Share one memory store across coding agents
• Trim model output verbosity via a local proxy

How you start.

1. 1Install the headroom CLI or the Python package
2. 2Wrap a coding agent so its traffic routes through a local proxy
3. 3Use the agent normally; cached originals stay locally retrievable

Before you begin.

• Requires Python 3.10+ and a machine where local processes can run; sandboxes blocking local processes cannot use it.
• Short conversational exchanges, prose and already-dense payloads see little or no token reduction.