Headroom
ActiveDescription
Context compression layer that compresses tool outputs, logs, files, and RAG chunks before they reach the LLM, saving 60-95% of tokens while preserving answer quality. Available as library, proxy, or MCP server.
Key Features
- Compress everything before LLM receives it, reducing 60-95% of tokens
- Three usage modes: library, proxy, and one-command agent wrapping
- Built-in MCP server with cross-agent shared memory
- ContentRouter auto-detects content type and selects optimal compressor
- CCR reversible compression: originals never deleted, retrieved on demand
- Compatible with Claude, Codex, Cursor, Aider, and Copilot agents
- headroom learn mines failed sessions and writes corrections back to CLAUDE.md / AGENTS.md
Use Cases
💡 Reduce API costs for AI agent workloads
💡 Compress code search, SRE debugging, and issue triage outputs
💡 Cross-agent context sharing and memory reuse
💡 Efficient transmission of long conversation history and RAG results
💡 Enterprise LLM application token optimization
Strengths & Limitations
✅ Strengths
- • Actively maintained, recent updates
- • High community interest (68.6k stars)
- • Permissive open-source license (Apache-2.0)
- • Responsive to issues, low backlog
Categories
Quick Start
# Install and wrap your agent
pip install "headroom-ai[all]"
headroom wrap claude # one-command wrap for Claude/Codex/Cursor
headroom proxy --port 8787 # or run as a proxy
headroom perf # inspect token savings