• •JetBrains introduced 'RTK' (Retrieval-based Token Knowledge), a method for Claude Code to query specific file chunks rather than scanning entire repositories.
• •Internal testing by JetBrains claims token consumption reductions between 60% and 90% in large-scale codebase interactions.
• •Community feedback on Hacker News remains limited, leaving it unclear how RTK performs under high-entropy code changes or extremely complex dependency structures.

JetBrains recently released data suggesting its RTK skill significantly lowers token usage for AI agents by selectively indexing repository content. This approach improves upon standard RAG (Retrieval-Augmented Generation) by optimizing how Claude Code interacts with massive file systems. However, while the efficiency gains appear high in controlled benchmarks, the methodology has yet to face rigorous public stress-testing outside of proprietary environments. Whether these savings translate to consistent performance in real-world, messy production environments remains a key variable for developers evaluating the cost-benefit of AI coding agents.