Markdown version of https://extraheadroom.com/docs/quickstart

# Quickstart

From download to your first measured savings in about ten minutes, most of which is a one-time model download. Four steps, one verification command, no configuration files.

## 1. Install the app

Download Headroom and open it: drag it into Applications on macOS, run the installer on Windows, or install the `.deb` (or run the AppImage) on Linux. Homebrew users on macOS can run `brew install --cask headroom` instead. Headroom lives in your menu bar or system tray; there's no window to keep open and nothing to remember to start. It keeps itself updated in the background on all three.

Two platform notes: the Windows installer is not code-signed yet, so SmartScreen warns the first time - More info, then Run anyway. The Linux builds are x86\_64 and built against glibc 2.39, so they need Ubuntu 24.04 or newer.

## 2. Let the first launch finish

On first run Headroom downloads its isolated Python runtime and a local compression model, roughly two gigabytes in total. This is a one-time cost; every later launch takes seconds. Everything runs on your machine: your prompts and code are never sent to our servers, and the model does its work locally.

## 3. Connect your coding agent

In the app, enable the connector for Claude Code, ChatGPT, or both. Headroom configures your shell so the agent routes through the local proxy. One thing to know: **terminals that were already open keep their old environment** , so open a fresh terminal before your next session.

## 4. Verify and run

In the new terminal, run `echo $ANTHROPIC_BASE_URL` (or `$OPENAI_BASE_URL` for ChatGPT). It should print an address on `127.0.0.1:6767`. Now work normally; the dashboard starts counting requests and savings from the first prompt. If the number stays at zero, the fix is almost always in [troubleshooting](/docs/troubleshooting), first entry.

## Where to go next

[How savings are measured](/docs/how-savings-are-measured) explains what the dashboard numbers mean (and what we refuse to count). [Add-ons](/docs/add-ons) covers the seven optional savers and why they're off by default. And [auto-learning](/docs/how-learning-works) is the part of Headroom that quietly compounds while you work.