Headroom Desktop

What it does: A cross‑platform desktop‑tray app that sits between your code‑generation tools (Claude Code, OpenAI Codex, OpenCode AI & Grok Build) and the LLM APIs they call. It runs a local optimisation pipeline that compresses prompts, tool output and logs, cutting the number of tokens sent to the model by roughly half while keeping the same results.

Why it matters: Token usage is the main cost driver for LLM‑powered coding assistants. By removing boiler‑plate, redundant JSON, and old conversation history before the request reaches the model, Headroom can halve the price you pay for Claude Code or Codex without any change to your workflow.

How it works:

• A Tauri‑based Rust tray app launches a local HTTP proxy on 127.0.0.1:6767.
• Your coding client is automatically re‑configured (via a tiny hook or environment variable) to point at that proxy.
• The proxy forwards the request to a managed Python runtime (headroom + optional tools) on a second local port, where the request is optimised and then sent on to the real Anthropic or OpenAI endpoint.
• Responses travel back the same way, with the original content restored on demand.
• The UI shows live dashboards of token savings, per‑client statistics and optional add‑on tools.

Key technical details:

• Zero host pollution – Headroom installs its own isolated Python (≈3 GB) under the app‑data folder; it never touches the system Python or global PATH.
• Cross‑platform – macOS (universal), Windows 10+, Linux (preview) builds are shipped from a single tag.
• Extensible – optional add‑ons (rtk, markitdown, serena, etc.) can be enabled from the UI; they all run locally.
• Open‑source shell – the desktop UI and proxy code are MIT‑licensed; the subscription service and backend are private.
• Safety – configuration changes are backed up, and on quit/uninstall all redirects are removed so the client works exactly as before.

Who should use it: Developers who already use Claude Code, Codex, OpenCode or Grok Build and want to reduce their token bills without rewriting their tooling. It also appeals to AI‑agent developers who need a predictable, locally‑hosted optimisation layer.

Installation: Homebrew cask for macOS or direct download of the DMG/EXE/AppImage from the GitHub releases page. The app is signed/notarised on macOS and includes a built‑in updater.

License: The desktop shell (src/ and src‑tauri/) is MIT‑licensed. Running the app requires a paid Headroom account (free 7‑day trial).