Headroom & RTK · AI Automation Society

Headroom & RTK

I’ve been tracking my Claude Code token usage over the last week using Headroom and RTK, and I found the combination surprisingly useful.

Headroom gives me the long-term picture across models, while RTK optimizes the prompts that actually get sent to Claude Code.

Check the umbers of my last 8 days:

* Headroom: 281M+ tokens saved (~$1.3k in proxy compression)

* RTK: 7.0M tokens saved from 9.0M input tokens (78% reduction)

One thing I learned: don’t judge these tools after one session.

RTK, for example, dropped below 50% savings for a while because I was working on tasks where there simply wasn’t much redundant context to compress. A few days later it climbed back to 78% as my workload changed.

So the percentage depends heavily on:

* the type of work you’re doing,

* how repetitive your prompts become,

* and how long you measure.

Looking at a single day can be misleading.

For me, the interesting part isn’t the exact percentage, it’s that over hundreds or thousands of Claude Code commands, the savings become significant.

The installation of both tools is pretty easy, if anybody want to give it a try:

RTK

Headroom

8 comments

Headroom & RTK