Headroom releases context compression layer for AI agents, reducing token usage by 60–95% while preserving accuracy

Headroom compresses tool outputs, logs, RAG chunks, files, and conversation history before they reach LLMs. It offers six compression algorithms (SmartCrusher for JSON, CodeCompressor for AST, Kompress-base for prose) and runs locally as a library, proxy, MCP server, or agent wrapper for Claude, Codex, Cursor, Aider, and Copilot.

On real agent workloads, compression achieves 92% token savings for code search (17,765 → 1,408 tokens) and SRE incident debugging (65,694 → 5,118 tokens), with 73% savings on GitHub issue triage and 47% on codebase exploration. Accuracy is preserved: GSM8K math benchmark maintained 0.870 baseline performance, and TruthfulQA improved from 0.530 to 0.560.

Headroom uses reversible compression (CCR), so original content is never deleted and LLMs can retrieve it on demand. It includes cross-agent memory for shared context across Claude, Codex, and Gemini, plus a `headroom learn` tool that mines failed sessions and writes corrections to agent documentation files.

Installation available via `pip install "headroom-ai[all]"` (Python 3.10+) or `npm install headroom-ai` (Node/TypeScript). Docker image available at `ghcr.io/chopratejas/headroom:latest`.