headroom

Repository: chopratejas/headroomDescription: Context compression layer for AI agents; reduces tokens by 60-95% before reaching the LLM. Key Features:

• SmartCrusher: Universal JSON compression for arrays, dicts, and mixed types.
• CodeCompressor: AST-aware compression for Python, JS, Go, Rust, Java, and C++.
• CacheAligner: Stabilizes prefixes to maximize provider KV cache hit rates.
• CCR (Reversible Compression): Originals are cached locally and retrievable via tools on demand.
• Cross-agent Memory: Shared store across Claude, Codex, and Gemini with auto-deduplication.
• headroom learn: Mines failed sessions to write corrections to CLAUDE.md / AGENTS.md.

Primary Use Cases:

• Reducing LLM costs by 60-95% without sacrificing accuracy.
• Handling massive code search results or long conversation histories.
• Sharing context and memory between different AI agents (e.g., Claude and Cursor).

Tags: #context-compression #token-optimization #ai-agent #memory Added: 2026-06-18 Source: GitHub

Notes / Why Notable

Headroom is a critical piece of infrastructure for high-volume AI agent workflows. It doesn't just truncate context; it intelligently "crushes" it while maintaining semantic integrity, and allows the LLM to retrieve the full original content if it needs specific details.