# Headroom

Aiolum story: https://aiolum.com/story/headroom/ · added 2026-10-25 · updated 2026-10-25

**In short.** Headroom (Desktop app that compresses the input of AI coding agents, Netherlands): US$2K monthly recurring revenue, October 2026. Evidence grade A: disclosed by a company or a filing, or verified by a payment processor. Team: 1 founder, Garm Lucassen, through his company Garm Tech B.V. (team size founder-stated). Launched 1 April 2026; still running, revenue growing.

Desktop app that compresses the input of AI coding agents · Netherlands · AI-native · solo founder · still running, revenue growing · group: Tools for using AI models

- Outcome: US$2K monthly recurring revenue, October 2026 (evidence grade of the headline number: A; headline in US dollars: 24300)
- Launch: 1 April 2026; months from launch to the outcome: 6
- Team: 1 founder, Garm Lucassen, through his company Garm Tech B.V. (team size founder-stated)
- Outside funding: None (bootstrapped, founder-stated)
- Main factor: Packaging
- Meters, 1 to 4: evidence strength 3, copyable pattern 2, capital needed 1
- Tags: AI-native, Solo founder, Bootstrapped, Running, B2B, Developer tools

**The product.** A menu bar app for macOS, Windows and Linux. It runs a proxy on the developer's own computer between Claude Code or ChatGPT Codex and the Anthropic or OpenAI servers, and it compresses tool output, logs, JSON and documents before each request leaves the machine. The originals stay on the computer, and the AI model can ask for them back. The app also installs optional add-ons and shows the savings and usage limits in a dashboard. After a 7-day free trial it costs from US$3 to US$40 a month, matched to the customer's Claude or ChatGPT plan.

**Why it worked.** The compression engine already existed as a free, open-source Python tool, but it was hard to install. Garm Lucassen packaged it into a desktop app that installs with a double click, connects the coding agents by itself and removes its changes when it quits. The engine's creator suggested that he build it and keep the Headroom name. Developers who hit the usage limits of their AI plans pay a small monthly price to make those limits last longer.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; calc = computed on this page from graded facts.

## Summary

Headroom is a desktop app for developers who use AI coding agents: programs such as Claude Code and ChatGPT Codex that read files, run commands and write code with a large language model. Every turn of such an agent sends text to the model, counted in tokens (pieces of words, the unit in which AI providers measure use and set limits). Most of that text is machine output: logs, search results, JSON. Headroom runs a local proxy, a small server on the developer's own computer that receives each request before it goes to Anthropic or OpenAI. The proxy shortens the noisy parts and forwards the smaller request. The company says this saves about 50% on the noisy items and typically 25% to 50% across a session.

Two parties built the product. The compression engine is the Headroom CLI, an open-source project created by Tejas Chopra and published under the Apache 2.0 license, a permissive license that allows anyone to use, change and sell the code if they keep its copyright and license notices. The desktop app is the work of Garm Lucassen and his Amsterdam company Garm Tech B.V. (Chamber of Commerce number 99602075). Its code is public on GitHub under the MIT license with the copyright line "Garm Tech BV". Lucassen wrote on AppSumo that he asked the engine's creator for an installable app, and that the creator told him to build it himself and to use the Headroom name. He shipped version 1 on 1 April 2026.

TrustMRR, a site that reads revenue directly from a startup's payment provider, shows US$2,025 in monthly recurring revenue (MRR, the subscription revenue that repeats every month, with a yearly plan counted as one twelfth of its price) from 222 active subscriptions on 25 October 2026. The payment provider is Polar. Verified revenue was US$2,540 in June 2026, fell to US$1,542 in August, and rose to US$2,284 in September and US$2,620 in the first 25 days of October. This page grades every number and shows four failed companies that sold developers a paid improvement to a tool owned by someone else.

## Pricing

Headroom prices on 25 October 2026, from the pricing section of the company's home page, with the AppSumo deal prices from the AppSumo listing

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Free trial | 0 | 0 | 7 days of use, no card |
| Pro | 3 | 4 | for Claude Pro or ChatGPT Plus; email support |
| Max x5 | 15 | 20 | for Claude Max x5 or ChatGPT Pro x5 |
| Max x20 | 30 | 40 | for Claude Max x20 or ChatGPT Pro x20; priority support |
| Team & Enterprise | custom | custom | shared controls, private deployment |
| AppSumo Lifetime Tier 1 | 29 one-time | 29 one-time | Pro level, 1 user license |

Each plan matches the customer's AI plan: Pro for Claude Pro or ChatGPT Plus, Max x5 for Claude Max x5 or ChatGPT Pro x5, Max x20 for Claude Max x20 or ChatGPT Pro x20. Yearly billing is 25% cheaper than monthly billing. The trial gives 7 days of use within 14 calendar days and needs no card; there is no free plan after it. At launch on 1 April 2026 the yearly prices were US$5, US$25 and US$50 a month and a free plan allowed use with 25% of the Claude plan; the current yearly prices are 40% lower (calculated). AppSumo sells one-time lifetime licenses at US$29, US$99 and US$199 for the three plan levels; AppSumo bills those sales itself.

## How much revenue does Headroom make? The numbers, graded

US$2,025 in monthly recurring revenue from 222 subscriptions in October 2026, six months after launch, read by TrustMRR from the payment provider Polar

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Monthly recurring revenue, 25 October 2026 | US$2,025 | A | TrustMRR snapshot; revenue last synced 25 October 2026 from Polar through an API key. This is the headline number. It is 1.25% above the site's floor of US$2,000 a month (calculated). | [1] |
| Active subscriptions, 25 October 2026 | 222 | A | TrustMRR snapshot from the payment records. | [1] |
| Average MRR per active subscription | about US$9.12 a month | calc | Calculated: 2,025 divided by 222 equals 9.12. The average is above the Pro price of US$3 to US$4 a month, so some subscribers are on the Max plans. | [1, 2] |
| Run rate from the MRR | US$24,300 a year | calc | Calculated: 2,025 times 12. | [1] |
| MRR growth in the last 30 days | +43.4% | A | TrustMRR snapshot. | [1] |
| Verified revenue, April and May 2026 | US$336 and US$538 | A | TrustMRR monthly timeline. March 2026 shows US$0. | [1] |
| Verified revenue, June 2026 | US$2,540 | A | 4.7 times May (calculated). The home page then offered 50% off all paid plans as founder pricing, and the app had added ChatGPT Codex. | [1, 14] |
| Verified revenue, July 2026 | US$2,251 | A | 11% below June (calculated). | [1] |
| Verified revenue, August 2026 | US$1,542 | A | 39% below June (calculated), and below the site's floor of US$2,000 a month. | [1] |
| Verified revenue, September 2026 | US$2,284 | A | 48% more than August (calculated). The founder-written profile says the AppSumo lifetime deal launched this month; AppSumo bills those sales itself. | [1, 5] |
| Verified revenue, 1 to 25 October 2026 | US$2,620 | A | A partial month, already 3% above June (calculated). At the same daily rate the full month would be about US$3,249 (calculated: 2,620 divided by 25, times 31). The daily table shows US$493 on 1 October; no source explains it. | [1] |
| Verified revenue, last 30 days to 25 October 2026 | US$3,198 | A | TrustMRR gives +80.2% against the 30 days before. A second snapshot on the same profile says US$3,240. The 30 daily values on the profile add up to US$3,199 (calculated). | [1] |
| Verified revenue, all time | US$12,111 | A | The monthly timeline from March to October 2026 adds up to the same US$12,111 (calculated). 55% of it came in the last 3 months (calculated: 6,601 divided by 12,111). | [1] |
| Payment provider | Polar | A | TrustMRR reads the revenue from Polar through an API key and marks it "Merchant of record: No". The terms of service name Polar as the payment processor for subscriptions. AppSumo purchases are billed by AppSumo under the terms, so they are probably not in these figures. | [1, 5] |
| Company behind the app | Garm Tech B.V., Amsterdam, Chamber of Commerce number 99602075 | C | Named in the terms of service and the privacy policy, and on garm.com with the same number. A B.V. is a Dutch private limited company. The number was not checked in the Dutch business register. | [5, 6, 10] |
| Founder | Garm Lucassen | C | He signs the AppSumo launch letter "I'm Garm, founder of Extra Headroom" and is the author of every post on the company blog. The TrustMRR founder handle is garmdotcom, the name of his company site garm.com. | [7, 11, 1, 10] |
| Team size | 1 person | C | Founder-written field on TrustMRR. The AppSumo listing gives the company size as the range 1 to 10. | [1, 7] |
| Outside funding | none stated | C | TrustMRR (founder-written) and the AppSumo listing both say Bootstrapped. | [1, 7] |
| Who built the compression engine | Tejas Chopra, Headroom CLI, Apache 2.0 | B | The company's own pages name him as the creator and say the app was built with his endorsement. The engine repository, headroomlabs-ai/headroom, was created on GitHub on 7 January 2026, carries the Apache 2.0 license with the notice "Copyright 2025 Headroom Contributors", and had 77,809 stars on 25 October 2026. | [3, 4, 9] |
| Who built the desktop app | Garm Tech B.V., MIT license | B | The repository gglucass/headroom-desktop was created on 30 March 2026; its license reads "Copyright (c) 2026 Garm Tech BV". It had 621 stars on 25 October 2026. The README says the shell is open source but the app needs a Headroom account and a paid plan to run. | [8] |
| Does the license allow selling the app? | yes | B | From the license text in the engine repository: Apache 2.0 allows commercial use and redistribution of the engine if the license and notices are kept, and it does not grant use of a trademark. The founder says the engine's creator asked him to use the Headroom name; the site credits him in every footer: "powered by Headroom by Tejas Chopra". | [9, 7, 2] |
| First version | 1 April 2026 | C | Founder's AppSumo letter. A Wayback copy of the home page from the same day shows a macOS download, Claude Code only, and paid plans from US$5 a month billed yearly. TrustMRR gives 27 March 2026 as the founding date. | [7, 13, 1] |
| Installs | 1,557 on 27 June 2026; 7,719 on 25 October 2026 | C | Counters on the home page. Growth of about 5 times in four months (calculated: 7,719 divided by 1,557). The founder-written TrustMRR profile estimates 7,527 users. | [14, 2, 1] |
| Google search impressions, last 30 days | 91,453 | A | Read by TrustMRR from Google Search Console. TrustMRR also shows 3,157 visitors in the last 30 days. | [1] |
| Measured input reduction, 183 users, 29 September to 5 October 2026 | 34% | C | Company blog post: 3.50 billion input tokens removed from a baseline of 10.30 billion. The output reduction of 33% is an estimate in the same post. | [12] |
| Listed for sale on TrustMRR | asking US$199,999 since 21 July 2026 | C | The asking price is set by the founder. The profile shows 4 offers received and a revenue multiple of 5.1. | [1] |
| Time from launch to the headline figure | 6 months | calc | Calculated: April 2026 to October 2026. | [7, 1] |

How to read this table: a grade A value was read by TrustMRR from the company's payment provider, Polar, or from Google Search Console. A grade B value is visible in public code or was stated by an independent party such as AppSumo. A grade C value is a statement by the company or the founder. A "calc" value was computed on this page from the rows above it.

## When did Headroom launch, and how did it grow? Timeline

From a first version on 1 April 2026 to verified recurring revenue just above US$2,000 a month in October 2026. Each entry carries its grade.

- **7 January 2026** (grade B [9]): The open-source compression engine, Headroom CLI, appears on GitHub. It is published on PyPI as headroom-ai under the Apache 2.0 license.
- **March 2026** (grade C [7, 8]): Garm Lucassen tries to install the engine, fails for an afternoon, and asks its creator for an app that installs with a double click. The creator tells him to build it and to keep the Headroom name. The desktop repository is created on 30 March.
- **1 April 2026** (grade C [7, 13]): Version 1 ships for macOS and Claude Code only. Plans: Free with 25% of the Claude plan, Pro at US$5, Max x5 at US$25 and Max x20 at US$50 a month billed yearly, after a 14-day trial.
- **April and May 2026** (grade A [1]): First verified revenue through Polar: US$336 in April and US$538 in May.
- **June 2026** (grade A [14, 1]): The app supports ChatGPT Codex, offers 50% off all paid plans as founder pricing and counts 1,557 installs. Verified revenue rises to US$2,540.
- **21 July 2026** (grade C [1]): The business is listed for sale on TrustMRR with an asking price of US$199,999.
- **August 2026** (grade A [1]): Verified revenue falls to US$1,542, the lowest month since May.
- **September 2026** (grade A [1, 7]): A lifetime deal launches on AppSumo at US$29 to US$199. Verified revenue through Polar rises to US$2,284.
- **5 October 2026** (grade C [5]): New terms of service: a 7-day trial without a card, no free plan after it, and rules for marketplace and lifetime purchases.
- **6 October 2026** (grade C [12]): The company publishes production data from 183 Claude Code users: input down 34% in one week.
- **25 October 2026** (grade A [1, 2]): TrustMRR shows US$2,025 in MRR from 222 active subscriptions and US$3,198 in revenue in the last 30 days. The home page counts 7,719 installs on macOS, Windows and Linux.

## Why did Headroom work?

### Packaging an existing engine

The hard part, the compression itself, was free code by another developer. Installing it needed Python, pip and changes to the agent's settings. The app installs its own Python runtime of about 3 GB, points Claude Code and Codex at the local proxy, and removes every change when the user quits. Customers pay for that installation, the updates and the dashboard, which the FAQ says plainly in its answer to "Isn't this just a paid wrapper around an open-source project?"

### Prices tied to the customer's AI plan

Each Headroom plan matches a Claude or ChatGPT plan: Pro for Claude Pro or ChatGPT Plus, Max x5 and Max x20 for the larger plans. The app reads the plan from the connected accounts. A heavy user who pays more for AI also pays more for Headroom, and the company prices each plan as "a small fraction of the subscription it stretches".

### Savings the user can see

The app shows the tokens saved in each session and an equivalent value at API prices. The company published a week of production data from 183 users and a blog post on how to state savings percentages honestly, and its terms warn that real savings can be far below the benchmarks.

### Search pages and marketplaces

The sitemap lists about 25 articles and blog posts on Claude Code and Codex costs and usage limits, and comparison pages with other token-saving tools. TrustMRR shows 91,453 Google search impressions in 30 days. The founder also lists Reddit, Hacker News and Product Hunt, and the app was sold as a lifetime deal on AppSumo from September 2026.

## How it was built

These details come from the public desktop repository, the company's pages and the founder-written part of the TrustMRR profile. The code details are grade B; the rest is grade C.

- A Tauri shell, a few megabytes in size, with a Rust intercept proxy on port 6767 and a managed Python backend on port 6768 that runs the Headroom engine and the rtk tool for shell output.
- The engine routes each block to a compressor by content type: SmartCrusher for JSON, an AST-based compressor for code, and Kompress-v2-base, a text compression model published on Hugging Face. Originals are stored locally so the model can retrieve them.
- Routing is set through an env block in the Claude Code settings file, a provider block in the Codex config file and a managed block in the shell profile, each with a backup before the edit.
- Optional add-ons from other open-source projects: RTK, MarkItDown, Serena, Codebase Memory, Context7, Ponytail, Caveman and Chisle. The README says only tools that run fully on the user's machine are included.
- Subscriptions through Polar; anonymous usage events through Aptabase, crash reports through Sentry and page statistics through Simple Analytics, according to the privacy policy. The founder lists React, Rust, Ruby on Rails and Python in the TrustMRR tech stack.
- Signed and notarized macOS builds; releases for macOS, Windows and Linux are published on GitHub from the same tag.

Sources: [8, 9, 6, 1]

## Who tried this before Headroom, and why did they fail?

Four companies in the Loot Drop database sold a paid improvement to developers or to users of a platform that someone else owned. The values below are from that database and are not independently verified.

The failed startups come from Loot Drop (https://www.loot-drop.io/), a free database of dead startups, retold here in our own words; Loot Drop describes its entries as AI-assisted summaries of public sources, so they carry no evidence grade.

### Kite DevTool (2014 to 2024; funding US$21M; cause: Competition)

- Promise: AI code completion inside the developer's editor, years before GitHub Copilot.
- Why it died: According to the database, it never turned free users into enough paying ones, and GitHub Copilot, built on a large language model and on GitHub's own distribution, took the market.
- Record: https://www.loot-drop.io/startup/2491-kite-devtool

### Fig DevTool (2020 to 2024; funding US$2.4M; cause: Unit Economics)

- Promise: Autocomplete and visual command builders for the terminal.
- Why it died: Developers used it daily but would not pay much for a terminal improvement, since free alternatives and their own scripts existed. AWS acquired it in 2024.
- Record: https://www.loot-drop.io/startup/2502-fig-devtool

### PostRocket (2011 to 2014; funding US$610K; cause: Product/Tech Failure)

- Promise: Better reach for Facebook pages through data on when to post.
- Why it died: Its value depended on Facebook's API and algorithm. When Facebook changed them and pushed its own advertising tools, the product lost its purpose.
- Record: https://www.loot-drop.io/startup/457-postrocket

### Cushion (2017 to 2025; funding US$15M; cause: Product/Tech Failure)

- Promise: Save consumers money by recovering bank fees automatically.
- Why it died: The database says banks saw it as taking revenue from them and blocked its access to accounts. A service that saves users money at a platform's cost depends on that platform's tolerance.
- Record: https://www.loot-drop.io/startup/2466-cushion

## What changed

Before AI coding agents, developers did not pay for model tokens in their daily work. Tools such as Kite and Fig improved the editor or the terminal, and the database records that developers would not pay much for that, because the gain was small and free options existed. There was nothing to save money on.

AI coding agents changed this. A Claude or ChatGPT plan has a usage limit per 5-hour window and per week; one reviewer quoted on the home page writes that on the US$20 a month Claude plan they occasionally hit both limits. The company's home page says that most of what uses up the limit is "the noise your tools pump into every prompt". A tool that cuts that output turns into extra hours of work on a plan the developer already pays for, so a small monthly price is easy to compare with the plan price. The remaining risk is the one PostRocket and Cushion faced: the product lives on top of services owned by Anthropic and OpenAI.

## Can one person copy Headroom now?

**Verdict: The pattern, yes, if the author of the engine agrees.**

Headroom resells no AI access and no closed code: it packages a permissive open-source engine, with credit and with the creator's agreement according to the founder. Anyone can do the same with the same engine, and Anthropic or OpenAI can build compression into their own agents.

1. Find a popular open-source tool that is hard to install for the people who would benefit from it, and read its license before you start.
2. Ask the author first, credit the author on every page, and keep the original free path visible.
3. Sell the parts the free tool lacks: one-click installation, automatic updates, clean removal and a dashboard that shows the benefit.
4. Price as a small share of the bill that your product reduces, in tiers that match the customer's existing plan.

**What is still open.** Many open-source AI tools have the same gap between strong code and hard installation: local speech-to-text, local document search, model routers for small teams. A signed desktop app with clean removal and a clear savings or time dashboard is open in each of them.

## What could go wrong for Headroom?

- **Just above the floor.** The MRR of US$2,025 is only 1.25% above the site's floor of US$2,000 a month, and verified revenue was below it in August 2026 at US$1,542. Revenue first passed US$2,000 in June 2026, so the record above the floor is about five months long, with one month below it.
- **Another person's engine and another company's platform.** The compression engine belongs to the open-source Headroom project of Tejas Chopra, and the app only works with Claude Code, Codex and a few other agents. If Anthropic or OpenAI build the same compression into their agents, change their usage limits, or restrict local proxies, the product loses its reason to exist. The terms of service forbid use that breaks Anthropic's or other providers' terms.
- **What the verified figures cover.** TrustMRR reads only the Polar account. AppSumo lifetime licenses, sold since September 2026 at US$29 to US$199, are billed by AppSumo and are probably not in these figures; they also bring no recurring revenue. The team size and the bootstrapped status are founder-written fields.
- **Savings claims.** The home page promises about 50% on noisy items and up to twice the usage. The company's own production week shows 34% less input, and its terms say that results "may be materially lower than any published benchmark" and that usage can even rise.
- **For sale.** The business has been listed for sale on TrustMRR since 21 July 2026 at an asking price of US$199,999, with 4 offers received. A new owner may change the product, the prices or the agreement with the engine's author.

## Sources

1. TrustMRR markdown profile: Headroom, revenue last synced 25 October 2026 — https://trustmrr.com/startup/headroom.md (grade A). Read the markdown profile live on 25 October 2026. The revenue, MRR, subscription and search figures come from Polar through an API key and from Google Search Console (grade A). The description, team size, funding, user estimate, channels, tech stack and asking price are founder-written (grade C).
2. Headroom home page (extraheadroom.com), read 25 October 2026 — https://extraheadroom.com/ (grade C). Read in full. Source for the pricing section, the install counter, the benchmark claims and the footer credit; checked for the purpose gate.
3. Headroom: What is Headroom?, read 25 October 2026 — https://extraheadroom.com/what-is-headroom (grade C). Read in full. Names the Headroom CLI as the engine, its Apache 2.0 license and its creator, Tejas Chopra.
4. Headroom FAQ, read 25 October 2026 — https://extraheadroom.com/faq (grade C). Read the answers on open source, the paid wrapper question, cost, licenses and reviews.
5. Headroom terms of service, last updated 5 October 2026 — https://extraheadroom.com/terms (grade C). Read in full. Names Garm Tech B.V. and its Chamber of Commerce number, Polar as payment processor, the trial, the marketplace and lifetime rules, and the savings disclaimers.
6. Headroom privacy policy, read 25 October 2026 — https://extraheadroom.com/privacy (grade C). Read the sections on data collected and third parties. Source for Polar, Aptabase, Sentry, Simple Analytics and Google Ads conversion tracking.
7. AppSumo: Headroom, with the founder's launch letter by Garm Lucassen — https://appsumo.com/products/headroom/ (grade B). Read in full. The deal prices, terms and review count are AppSumo's (grade B). The launch letter, the founding date and the company details are the founder's (grade C).
8. GitHub: gglucass/headroom-desktop, the desktop app's repository — https://github.com/gglucass/headroom-desktop (grade B). Read the README and the LICENSE file through raw.githubusercontent.com, and the creation date, stars and license through the GitHub API, on 25 October 2026.
9. GitHub: headroomlabs-ai/headroom, the open-source compression engine — https://github.com/headroomlabs-ai/headroom (grade B). Read the README, the LICENSE and the NOTICE file through raw.githubusercontent.com, and the creation date, stars and license through the GitHub API, on 25 October 2026.
10. Garm.com, the site of Garm Tech B.V. — https://garm.com/ (grade C). Read in full. Lists the company's products, among them Extra Headroom, Evalship, Nommer and Criticaster, with the same Chamber of Commerce number.
11. Extra Headroom blog index, read 25 October 2026 — https://extraheadroom.com/blog (grade C). Read in full. Every post from June to October 2026 is signed by Garm Lucassen.
12. Extra Headroom blog, Garm Lucassen, 6 October 2026: Extra Headroom in production: Input -34%, Output -33% across 183 Claude Code users — https://extraheadroom.com/blog/claude-code-savings-real-usage (grade C). Read the results and method sections.
13. Wayback Machine: extraheadroom.com on 1 April 2026 — https://web.archive.org/web/20260401150310/https://extraheadroom.com/ (grade C). Read in full. The launch-day page with the first plans and prices.
14. Wayback Machine: extraheadroom.com on 27 June 2026 — https://web.archive.org/web/20260627130049/https://extraheadroom.com/ (grade C). Read in full. Source for the June install counter, Codex support and the founder pricing.
15. Evalship home page, read 25 October 2026 — https://evalship.com/ (grade C). Read the first screen. Another Garm Tech product (quality checks for LLM evaluations), checked for the purpose gate.
16. Nommer home page, read 25 October 2026 — https://nommer.ai/ (grade C). Read the first screen. Another Garm Tech product (a recipe split between two phones), checked for the purpose gate.
17. Criticaster home page, read 25 October 2026 — https://criticaster.com/ (grade C). Read the first screen. Another Garm Tech product (one score from expert product reviews), checked for the purpose gate.
18. Loot Drop database export, 1,749 failed startups — https://www.loot-drop.io/ (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.