Headroom is a macOS menu bar app that reversibly compresses tool output, logs, and boilerplate before sending prompts to Claude Code and Codex, cutting token costs by ~50% with no loss in model output quality.
What is Headroom?
Headroom is a local proxy application that intercepts prompts sent to Claude Code and OpenAI Codex, compressing noisy elements like build logs, JSON blobs, and shell output into a compact format while keeping the original content retrievable on demand. It runs on macOS (Intel and Apple Silicon) and is built on the open‑source Headroom CLI by Tejas Chopra, with his endorsement and support. The desktop app adds a signed notarized installer, automatic updates, a menu‑bar UI, and usage statistics — the CLI itself is free and self‑hosted.
Key Features
- Reversible compression — Compresses logs, HTML, JSON, and shell output; the model can request the original content on demand, preserving access to all information.
- ~50% token savings — Benchmarks show savings from 47% (codebase exploration) to 92% (code search with 100 results); realistic multi‑tool agent tasks save 61%.
- Privacy‑first local processing — All optimization runs on your machine; prompts and code never leave your computer.
- No quality degradation — QA benchmarks show identical or improved F1 scores (e.g., +0.02 F1 on HTML extraction, +2% exact match on SQuAD v2/HotpotQA) versus uncompressed baselines.
- Menu bar dashboard — Real‑time view of token savings, activity history, project learnings written to agent memory, and one‑click add‑ons.
- Add‑on marketplace — One‑click install of RTK (Reversible Tokenizer), MarkItDown, Ponytail, and other optimisations.
- Works with Claude Code and Codex — Supports any Claude Pro, Max, or ChatGPT Plus/Pro plan, including Max tiers.
Who is it for?
- Solo developers using Claude Code or Codex daily — Stretch your current plan to cover roughly twice as many sessions after a single install.
- Engineering teams on shared AI subscriptions — Reduce per‑developer token waste from logs and boilerplate without upgrading to a higher tier.
- Teams that want CLI automation — The free open‑source CLI (same engine) can be wired into CI/CD pipelines for cost‑efficient agent workflows.
What can you do with Headroom?
- Extend multi‑tool debugging sessions — Investigate a memory leak with a 4‑tool agent; Headroom compresses logs and output, saving 61% of tokens while the agent reaches identical conclusions.
- Save on code‑search‑heavy workflows — Search a large codebase returning 100 results; compression reduces token usage by 92%.
- Combine with other optimisations — Chain Headroom with RTK and add‑ons to triple effective usage, as reported by users.
How does Headroom work?
- Install the macOS app (signed
.dmg) and grant necessary permissions.
- The app runs as a local proxy, intercepting all prompts to Claude Code or Codex.
- It reversibly compresses logs, HTML, JSON, shell output, and boilerplate before forwarding the prompt.
- The model sees compressed context but can retrieve the original content on demand.
- The menu bar dashboard shows real‑time token savings, activity history, and project learnings written to agent memory.
Pricing
Headroom is freemium. A 7‑day free trial requires no credit card. After trial, plans start at $3/month (billed annually) for Claude Pro / ChatGPT Plus, $12/month for Max ×5, $24/month for Max ×20. Founder pricing offers 40% off all paid plans (limited spots). Team & Enterprise plans are custom (contact via website). The underlying Headroom CLI remains free and open‑source.
FAQ
Is Headroom free?
The desktop app has a 7‑day free trial; paid plans start at $3/month. The CLI version is free and self‑hosted. See the pricing page for full details.
