Headroom docs
Headroom is set-and-forget by design, which means most of what it does happens quietly. These guides make the quiet parts visible: what's running, what it's saving, what it's learning, and which extras are worth switching on.
Getting started
Quickstart takes you from download to your first measured savings in about ten minutes: install, connect Claude Code or Codex, verify the routing with one command, and read the dashboard.
Guides
How savings are measured: the dashboard's number is deliberately conservative. What it counts, what it refuses to count, and why you can multiply it against your own bill.
How auto-learning works: Headroom notices patterns in your sessions and teaches them to your agent. Here's why the first learnings take a few sessions to appear.
Add-ons, and why they're off by default: seven of them, from RTK's quieter terminals to Serena's symbol-level code reads. Each saves real tokens, but unlike the core proxy they change how your agent behaves, so they're one click away instead of on by default.
Beyond token savings: on a subscription you don't pay per token, so what does compression buy you? More usage inside your rate limits, for a start.
When something looks off
Troubleshooting walks the five symptoms we actually see (zero savings, slow first launch, no learnings, missing Remote Control, a busy port), each with its fast check.
Product
The changelog is the running log of what's shipped, newest first. And uninstalling is documented end to end, every leftover listed, because trust includes the exit.
Beyond the docs
Elsewhere on the site: the FAQ covers privacy, platform support, and pricing, and the blog goes deep on token economics.
See it on your own traffic: install Headroom and run a normal session.