RTK is a high-performance CLI proxy written in Rust that reduces LLM token consumption by 60-90% on common developer commands. It operates as a single binary with zero dependencies and is designed to integrate seamlessly with AI coding agents like Claude Code, GitHub Copilot, Cursor, and 11 other supported tools. The core functionality filters and compresses command outputs before they reach an LLM's context window, applying four optimization strategies: smart filtering to remove noise like comments and whitespace, grouping to aggregate similar items, truncation to preserve relevant context while cutting redundancy, and deduplication to collapse repeated log lines with counts.
The repository demonstrates substantial real-world impact through documented token savings. In a typical 30-minute Claude Code session on a medium-sized TypeScript or Rust project, RTK achieves approximately 80% total token reduction across common operations. Individual commands show dramatic improvements: git operations save 75-92% of tokens, test runners achieve 90% savings, file listing and grep operations save 80%, and build tool output like cargo test and npm test see 90% reductions. The tool supports over 100 commands across multiple categories including file operations, git workflows, GitHub CLI, test runners, build and lint tools, package managers, AWS CLI, container tools, and infrastructure-as-code utilities.
RTK's most effective deployment method is through an auto-rewrite hook that transparently intercepts Bash commands and rewrites them to rtk equivalents before execution, achieving 100% adoption across conversations and subagents with zero token overhead. The hook system works natively on Windows as of version 0.37.2, running as a native binary command rather than requiring Unix shell dependencies. Installation is straightforward through Homebrew, cargo, or pre-built binaries for macOS, Linux, and Windows platforms.
The repository overlaps with contributors from dbt-labs/dbt-core, anthropics/claude-code, and other projects, suggesting integration within a broader ecosystem of developer tools.
Configuration is managed through a TOML file located at ~/.config/rtk/config.toml on Linux and macOS, with macOS also supporting the standard Application Support directory. When commands fail, RTK preserves full unfiltered output so LLMs can read results without re-execution. The project is licensed under Apache 2.0 and maintains comprehensive documentation including a full user guide at rtk-ai.app/guide, architecture documentation, contribution guidelines, and security policies. The repository supports multiple languages in its README, including French, Chinese, Japanese, Korean, Spanish, and Portuguese, reflecting its international user base.
