Side by side
| Headroom CLI | Headroom for desktop | |
|---|---|---|
| What it is | Open-source Python proxy and library (Apache 2.0). | macOS menu bar app built on the CLI. |
| Price | Free. | Paid subscription, 7-day free trial. |
| Install & setup | pip install, then configure your agent's base URL and keep the proxy running yourself. | Signed, notarized installer; the app manages the runtime, configuration, and updates. |
| Updates | Manual — you pull new versions and restart. | Automatic. |
| Visibility | Logs and CLI output. | Menu-bar dashboard with live token savings and activity. |
| Add-ons | Bring your own. | One-click Ponytail, RTK, and MarkItDown. |
| Support | Community (GitHub, Discord). | Ongoing support from the app team. |
Same engine, two packages
The desktop app is based on the open-source Headroom project created by Tejas Chopra, and is built with his endorsement and support — we also contribute improvements back upstream. Both do the core job the same way: they sit between your coding agent and the model as a local proxy, reversibly compress the bloated tool output, logs, and boilerplate in your context, and let the model pull back any original on demand. The compression quality is the shared engine; nothing about the paid app makes the CLI a lesser version of that.
What you pay for in Headroom for desktop is everything around the engine: installation without touching a terminal config, a managed runtime that stays up and stays current, a dashboard that shows what you're actually saving, and the add-on stack — Ponytail, RTK, and MarkItDown — enabled from one screen.
Which should you use?
Use the CLI if you're comfortable managing a Python environment and a long-running proxy, want Apache-2.0 code you can read and modify, or are integrating compression into your own tooling. Use the app if you'd rather spend those hours on your own work: it is set up in about two minutes and roughly doubles the usage of the Claude or ChatGPT plan you already pay for. The install guide walks through both paths in detail.