Guides

Practical notes on cutting the tokens AI coding agents spend.

• Where coding-agent tokens go, and the tool for eachCoding agents spend tokens in five places: shell output, file reads, tool and MCP results, conversation history and replies. Which token-reduction tool fits each: RTK, SuperCompress, Headroom, Context Mode, NeuralMind, /compact and Caveman.
• Headroom vs Caveman vs SuperCompressHeadroom, Caveman and SuperCompress all cut coding-agent tokens, but in different places. Headroom compresses context locally, Caveman shortens replies, and SuperCompress keeps only the lines that answer the question, hosted. How to choose.