RTK

RTK is a high-performance command-line proxy that reduces the shell output your AI coding agent reads—cutting common development-command output by 60–90%. It filters, groups, truncates, and deduplicates verbose terminal output before it enters your LLM context, helping agents stay focused while reducing token usage.

Built as a single Rust binary with zero runtime dependencies, RTK supports 100+ common commands and adds less than 10 ms of overhead. It works transparently with popular AI coding tools, rewriting commands such as git status, cargo test, or pytest into compact RTK equivalents.

• Features

Token-aware output compression — Removes noise, boilerplate, repeated lines, and irrelevant details from shell output

Smart filtering strategies — Applies filtering, grouping, truncation, and deduplication based on command type

Broad command support — Optimizes files, Git, GitHub CLI, test runners, linters, builds, package managers, containers, cloud tools, and more

Automatic command rewriting — Hooks intercept Bash commands and route them through RTK without requiring the agent to call it explicitly

Failure recovery — Saves full unfiltered output for failed or truncated commands so agents can recall details without rerunning them

Savings analytics — Track estimated reductions with rtk gain, daily history, graphs, and JSON exports

Wide agent compatibility — Supports Claude Code, GitHub Copilot, Cursor, Gemini CLI, Codex, Windsurf, Cline, OpenCode, and more

Cross-platform — Native support for macOS, Linux, Windows, and WSL

Privacy-conscious — Telemetry is disabled by default and requires explicit opt-in

Note: RTK reduces command output, not necessarily your total AI bill. Token savings depend on how much of your context consists of shell output; prompt, system context, conversation history, and output tokens also affect cost.