RTK Just Hit 30K Stars Because Everyone's Token Bill Is Out of Control

A single Rust binary is saving agentic coding users 60-90% on token costs, and it took about five minutes to set up.

RTK (Rust Token Killer) released v0.37.1 on April 18 and sits at 30,500 GitHub stars. The tool acts as a CLI proxy between your AI coding assistant and shell commands. Every time Claude Code or Cursor runs git status, ls, find, or any of 100+ supported commands, RTK intercepts the output, applies smart filtering, grouping, truncation, and deduplication, then passes a compressed version to the LLM context. A git status that normally produces ~200 tokens gets squeezed to ~20. Overhead is under 10 milliseconds. Zero dependencies.

I'm writing about this the same week Uber publicly admitted their Anthropic bill spiraled out of control (more on that below). The timing isn't coincidental. Token costs are the silent tax on every agentic coding workflow, and most developers don't realize how much context budget they're wasting on verbose command output that the model barely needs. RTK doesn't change how you work. It just makes every tool call cheaper.

The architecture is dead simple, which is why it works. RTK doesn't modify your AI tool. It doesn't require API key changes or config rewrites. You prefix your commands with rtk or set it as your shell wrapper, and it handles compression transparently. The Rust implementation means it's fast enough that you genuinely can't feel it. I've been running Claude Code for hours-long sessions where the cumulative token waste from raw git diff output alone was probably costing me $5-10 per session. RTK claims 60-90% reduction across typical workflows. Even if the real number is 40% for your usage pattern, that's meaningful money if you're shipping code daily.

What I find interesting is the star trajectory. 30K stars for a CLI proxy. That tells you the pain is real and widespread. This isn't a cool demo repo people star and forget. It's a tool people install and use because they can feel the difference in their billing dashboard.

The connection to the broader token economics story is obvious. As models get more capable and context windows get larger, the temptation is to stuff more into every request. RTK works against that impulse at the infrastructure level. Smart compression before the context window, not smarter prompting inside it.

What to do now: Install RTK (cargo install rtk or grab the binary from the releases page). Run your normal coding workflow for a day, then compare token usage. If you're using Claude Code or Cursor with any regularity, this pays for itself in the first session.