Headroom

Context compression for AI agents — reduce the tokens sent to an LLM by 60–95% while preserving meaning, by compressing tool outputs, logs, search results, RAG chunks, files, and conversation history before they reach the model. It auto-detects content type (JSON / code / text) and supports CCR (Compress-Cache-Retrieve): originals are cached locally and fetched on demand.

Upstream: headroom by chopratejas (Apache-2.0); published on PyPI as headroom-ai.

When to Use

• A tool/command returns a large dump (verbose logs, big JSON, long file listings, search results, large files) that would bloat the context window.
• The user asks to compress context, shrink a payload, or save tokens.
• Compress before the large content is committed to context.

Setup