What's new in Headroom.

Every release, newest first. Updates download on their own and install the next time Headroom opens.

  • Latest0.9.87
  • 86 releasessince 30 September 2026
Latest
  • Headroom notices a new Claude plan. When Claude Code (or a tool that runs it, like T3 Code) is signed in to a Claude account Headroom doesn't have, the menu offers it by name with one click to add it.
  • Adding a Claude account no longer says "Signed in again" when your browser picked an account that was already here. It tells you nothing was added and how to reach the other account, and you can copy the sign-in link into a private window.
  • With several Claude Code setups on one Mac, the menu bar follows the account you used last, and an AI agent that asks Headroom is told the account of its own session.
  • The yellow model tile names what's in it ("8 more models") and lists the biggest ones with their shares, even in a smaller window. Click a row to see that model; hover it for its tokens.
  • "+5 more" opens the full list with each model's tokens and share, and the model you picked stays marked on the tile.
  • The share window now has a close button, top right. Esc still closes it too.
  • On a share card where one model does most of the work, the small "Other" tile lists its models in clean rows instead of a cramped line of small print.
  • On your share card and in the share video, the "Other" tile now says what is in it: each model by name with its share, instead of just a count. A model you used almost as much as your top two no longer disappears into "Other".
  • When your top models are close, the leading tile is wide enough for its figure to read as the biggest.
  • In the share video, the big number no longer looks cut off as it arrives. It now fades up whole, starts counting sooner, and its digits keep clean edges once it settles.
  • The models on your share card and in the share video are now Headroom's real mosaic: the biggest model's tile wraps round its neighbour, exactly like the dashboard, instead of three plain boxes.
  • The square card gives the mosaic more room, and the top model's share is always the largest figure in it.
  • The share video is rebuilt from scratch. It opens on your headline, then your tokens roll up on a counter while the bars rise day by day, your models take over the whole frame and grow into their real shares, your facts land one by one, and it settles on your card. About twelve seconds, in all three sizes and looks.
  • A new share card to match. Your models now appear as Headroom's mosaic (the biggest in cobalt, the next in ivory, the rest in gold), beside the period's bars with the busiest one called out by name.
  • The big number no longer gets its last letter clipped in the preview, and its digits sit evenly whatever the total.
  • Account and project names, when you turn them on, now sit neatly under the date. They stay off unless you choose them.
  • Claude Code sessions run from other apps that keep their own Claude folder, such as T3 Code, now count in your usage, projects and API value.
  • Prices for GPT-6.1 Terra, Claude Fable 5.2, Mythos 5.2 and Haiku 5.5, so their API value is no longer left out. Fable 5.2's fast mode counts at double, like Opus 5.5.
  • Model limits such as Claude's weekly Fable limit now show pace, the even-pace tick and how long they last, like the 5-hour and weekly limits. Fable has its own allowance and can run out before your weekly does.
  • AI agents now see Codex sessions that aren't tied to one Codex account as Codex, instead of an unknown tool.
  • Right after Headroom opens, AI agents now get your usage split by project as soon as your history has loaded, instead of a few minutes later.
  • Ask your AI about your AI usage. Headroom now works with Claude Code, Codex, Cursor and the Claude app: connect them in Settings › AI agents, then ask "how much Claude do I have left?", "what did each project cost this month?" or "does this job fit in my 5-hour window?".
  • Your agent gets three read-only tools: limits left on every account (with resets and pace), tokens and API value for the last 30 days by tool, account, model, project or day, and API value per project.
  • Claude Code and Codex also get Headroom's skill, which tells the agent when to check your limits and how to read API value. Everything stays on your Mac until your agent reads it. More at headroom.bar/mcp.
  • Claude Code history works again for heavy users. Headroom stopped reading Claude Code's history once you passed 100,000 requests in a month; it now holds 400,000 and, past that, keeps the newest instead of stopping. Your dashboard and Projects catch up on their own after the update.
  • The menu bar follows the Claude account you're working in. On automatic, Claude Code's reading now shows the account Claude Code is signed in to, the way Codex follows its sign-in, instead of whichever account had the freshest reading.
  • A folded account in the menu now shows its weekly limit, not the 5-hour. The weekly is the one that decides your week, so "21 ahead" on a 5-hour that resets in minutes no longer hides "9 behind" on the weekly. The 5-hour only takes over when it's about to run out before it resets.
  • Cloud sessions with several repos attached can now be counted. Claude Code doesn't run a repo's hooks in those sessions, so Settings › Connections › Cloud sessions has a short setup script to paste into your cloud environment once. Each repo then gets its own work, and requests that don't name a file count where the session was working.
  • Update the hook in each repo from Settings › Connections (one commit each). The new hook also works when a session starts above your repos.
  • Adding Headroom to a repo now changes only Headroom's own lines in .claude/settings.json, so your diff stays small.
  • The commit command works in repos whose .gitignore only allows listed files, and the "Commit and push it" step goes away once you've pushed.
  • A new way to share your activity. One big number up front, then your API value, how much input was reused, your busiest hour or day, the period's rhythm as bars with the peak in gold, and which models did the work. Everything uses Headroom's own enamel look.
  • Three sizes and three looks. Square for X, Instagram and LinkedIn, Landscape for X and LinkedIn link previews, and Story for phones. Each comes in Dark, Light or Cobalt.
  • Share videos that are made to be posted. A 12-second clip: your tokens count up while the bars build, your models take a bow, then it settles on your card. Comes in all three sizes, readable with the sound off.
  • A bigger, clearer share sheet. A large live preview, headline ideas written from your numbers, one-tap choices for what to show, and Share… for Messages, Mail and AirDrop, plus Copy, Save and Post on X. It remembers your choices.
  • Project and account names still stay off unless you turn them on. Prompts, code and session IDs never appear.
  • Your totals, front and centre. The dashboard now shows how many tokens you've used and their API value in big numbers at the top of the activity panel, and in the line under the greeting, so they're in every screenshot. They follow the range you pick: last hour, today, 7 days or 30 days.
  • Every account's limits at a glance. On All accounts, the bottom card lists each account's limit, most at risk first: how much is left, whether you're ahead or behind an even pace, and when it resets. Click one for its full pace.
  • "API value" is now the one name for what your tokens would cost on the pay-as-you-go API.
  • Cloud sessions load much faster. Headroom reads all your Claude Code cloud sessions in one go instead of one at a time, so a repo with hundreds of sessions no longer slows your Mac every couple of minutes.
  • Cloud sessions show up within minutes. Headroom now checks for Claude Code cloud work every two minutes instead of ten. Update your repo's cloud tracking in Settings › Connections › Claude Code (one commit) and a long reply shows while it's still running, not only once it finishes.
  • Searches that named a single file no longer show up as a project of their own.
  • Send feedback's report says when cloud work was last checked, and for each project folder whether your sessions edited it or only read it.
  • Your account keeps up with the Projects page. When Headroom files work under a different project, as this week's fixes do, your account's copy and your other Macs now follow, instead of keeping the old split.
  • Send feedback's report now explains how the Projects page filed your Claude Code work in repos with several projects, so a wrong split can be fixed from one report.
  • Projects inside one big repo count properly. When a session looks at the top of the repo (a cd, a git status, a search across everything), that no longer pulls its work onto the repo itself; it stays with the project the session works on. Headroom reads your history again once to apply it, so the Projects page settles after a few minutes.
  • MiniMax. Paste a MiniMax API key in Settings › Connections. A Token Plan key shows your 5-hour and weekly limits for MiniMax's coding models; a pay-as-you-go key shows what's left in your balance.
  • Command Code. Signed in to Command Code on this Mac? Headroom reads your 5-hour and weekly limits and your credits, the same numbers its /usage shows.
  • Sessions on your Mac count where the work happened too. When a Claude Code session started at the top of a repo reads and plans before it opens a project's files, that time now counts toward the project it works on. Headroom reads your Claude history again once to apply it, so the Projects page settles after a few minutes.
  • Cloud sessions count where the work happened. When a Claude Code cloud session reads and plans before it opens a project's files, that time now counts toward the project it works on, not the main repo. Your Projects page updates within a few minutes, history included.
  • Send feedback never makes you wait. Press Send straight away: Headroom adds the diagnostic report as soon as it's ready, and you can close the window.
  • Send feedback never makes you wait. Press Send straight away: Headroom adds the diagnostic report as soon as it's ready, and you can close the window.
  • Steadier Claude limits. Claude was turning down about half of Headroom's checks, so your numbers sometimes went stale. Headroom now paces its checks the way Claude accepts, and your limits keep updating.
  • Fewer "Claude asked Headroom to wait" moments. Claude was turning down about half of Headroom's checks; now Headroom paces them the way Claude accepts, and refreshing yourself still checks right away.
  • Headroom now uses a Headroom account. Setup asks you to sign in (GitHub, Google, Apple or an emailed code) right after the welcome, and if you already use Headroom without one, the menu and dashboard ask until you do. Your account keeps your beta access and Founding licence with you, and adds up every Mac you sign in on.
  • Setup and Settings › Account say plainly what your account holds: your email and name, your Macs, your synced usage (tokens, models, costs and project names), your tools' limits and plans, and daily usage stats. Your prompts, replies, code and files never leave your Mac.
  • Usage stats are now sent with your account, so feedback about what breaks reaches the right person. See what's sent and the switch to turn them off are in Settings › General.
  • Service status for Codex now covers what Codex runs on. Codex lives inside ChatGPT and uses OpenAI's Responses API, so Headroom also tells you when OpenAI reports a problem with ChatGPT or the Responses API, not just with the Codex group. Unrelated parts, like the Images or Audio APIs, still never mark Codex.
  • Clearer status wording. Each part is named in full ("Responses API", "ChatGPT Conversations"), and when several are affected the notice names the provider ("OpenAI: degraded performance") and lists them.
  • A much smoother dashboard with a big history. Switching between 1h, 5h, Today, 7d and 30d now draws about three times as many frames: rolling numbers render on the graphics card, the history chart fades to the new range instead of stretching every bar, and the hidden Projects page no longer redraws alongside the one you're looking at.
  • Clicks and trackpad taps answer sooner, since they no longer wait behind slow frames.
  • Smoother with a big history. When new requests come in, Headroom now compares and replaces your history in the background instead of while it draws, so numbers animate cleanly and the menu opens without a hitch, even with a million requests on record.
  • Buttons show their press when you tap the trackpad, not just when you click it.
  • How long your quota lasts, right in the menu. Under each limit: "Runs out in 2d 7h" when your pace this cycle uses it up before the reset, or "Lasts ~5d 6h at this pace" when it reaches past it. The dashboard's capacity card now says how long it would last too, not just that you'll make it.
  • Clearer service status. The notice names exactly what's affected (for example Codex in ChatGPT Desktop, not Codex as a whole), says where the provider is with it ("OpenAI is monitoring a fix"), and shows when Headroom last checked, so a long-running incident never looks like a stale reading.
  • Your first sync shows its progress. Signing in to a Headroom account sends your recent history up once, which can take a few minutes, so Settings › Account now says how far it's got ("Sending your history · 72,400 of 181,250 requests"), lists this Mac straight away, and sends the newest days first so your other devices see today soonest.
  • Service status. When OpenAI, Anthropic, GitHub, Cursor or Windsurf report a problem with a tool you use, its logo in the menu bar gets a small dot, and the menu opens with what's wrong, since when, and a link to their status page. Only the parts each tool uses count, so a ChatGPT image outage never marks Codex. Turn it off in Settings › Menu bar.
  • The menu bar's capacity bar sits right beside its reading, centred on it. Before, macOS placed them apart and the bar sat a little low.
  • New: Capacity bar only, in Settings › Menu bar › Indicator. Just the bar, no number. With Show limit set to Both, it's two thin bars, weekly above 5-hour. The reading is still in the tooltip.
  • The menu always fits your screen. When an update note or a notice takes room, your accounts scroll instead of running off the bottom, and this note is now a few short lines with Headroom's own icon.
  • Amp shows your plan, credits and daily spend. Current Amp keeps no token counts on your Mac, so Headroom asks the installed Amp for the same report amp usage shows, at most every 10 minutes.
  • Kimi Code keeps its last limits reading. Kimi renews its sign-in only while it runs, so between sessions Headroom shows the last reading with its time instead of an error, and an account without a Kimi Code plan says so.
  • Eight more tools. Mistral Vibe, Kimi Code, Qwen Code, Factory Droid and Amp: token history read from their own logs on this Mac, by model and project, with what it would cost at API prices. Kimi Code also shows its 5-hour and weekly limits, using the sign-in Kimi Code already keeps. Kiro shows this month's credits, as Kiro last checked them.
  • OpenAI API and Anthropic API spend. Paste an organization admin key and Headroom shows what each provider charged per model and day, with your tokens, and OpenAI's monthly spend limit if you set one. Admin keys can only read usage. Anthropic doesn't give individual accounts one.
  • Turn on only what you use in Settings › Connections. Setup now lists the tools on this Mac, plus a count of the rest.
  • Beta stats now say whether each tool worked that day, as a short fixed code like "limits not found", never the error itself. It's how we spot a tool update that breaks Headroom before anyone has to tell us. See exactly what's sent in Settings › General › See what's sent, or turn beta stats off there.
  • Codex sign-in problems are recorded by name, so Send feedback reports are clearer.
  • Codex connects again for people who only have the ChatGPT app. Its latest version keeps Codex in a new place, and Headroom said Codex wasn't installed. Headroom now asks macOS where the app is and finds Codex inside it, wherever OpenAI puts it next.
  • Settings › Connections › Codex shows which Codex Headroom uses, with Locate Codex… if it ever can't find one.
  • Window appearance: switching back to System follows your Mac again straight away. It could stay stuck in Light or Dark until you picked the other one first.
  • Settings › Spend shows the models you actually use, Claude included, with each one's API price, plus every published price Headroom tracks. It used to list seven OpenAI models.
  • Signing in with Google, GitHub or Apple no longer crashes Headroom as the browser sends you back.
  • A model with a small share gets a readable tile instead of a thin sliver or a stray line under the others. The percentages still show its exact share.
  • Finishing setup always lands somewhere. If Headroom's menu bar icon is hidden (behind the notch, in a crowded menu bar or a menu bar manager), "Show me the menu" now opens the dashboard instead of closing setup with nothing on screen.
  • Clicking Headroom in the Dock after setup opens the dashboard. It could open setup again.
  • At the end of setup, "What's sent" shows the report your Mac will actually send. It was showing a sample report (an older version, tools you may not use) instead of yours.
  • Models, the hourly chart and the compact window show your hours again on 1 hour, 5 hours and Today if you don't use Codex. They were asking you to "include your Codex accounts" even when there were none to include.
  • Headroom accounts. Sign in under Settings › Account with an emailed code or GitHub, and every computer you sign in on adds up: your laptop now shows the usage of the Mac your agents run on, in totals, Projects and history. Codex from another device counts as Codex; Claude Code and the other tools show each device as its own account.
  • Only usage numbers sync: token counts, models, times, costs and project names. Never prompts, replies, code or files. Signed out, Headroom works exactly as before.
  • Settings › Account lists your devices, lets you remove one you've stopped using, and deletes the account and everything synced to it if you want.
  • Codex history shows up on day one. Without CodexBar installed, setup used to start Codex's daily history from the moment you installed Headroom; it now reads the last 35 days of your Codex logs, a little at a time in the background.
  • A Codex turn that leaves out one of its token counts is still counted, as zero for the missing part, instead of being skipped.
  • Monthly limits are labelled "Monthly limit" in an account's details, not "720-hour limit".
  • Small copy fixes: "Add an OpenCode Go key", the welcome screen counts all eleven tools, the Projects empty state no longer names only Codex, and the project grouping search says "Find a source".
  • Grok Build is here. Headroom shows your Grok usage pool, the one SuperGrok and X Premium+ share across Chat, Build and the API, with when it resets, by asking the installed Grok Build, the same reading its own /usage shows. It never reads your Grok sign-in.
  • Grok Build's history shows up in Projects, filed under the repo and project folder each turn worked in, with what xAI actually charged rather than an estimate.
  • Claude Code cloud sessions can show up in Projects. Add Headroom to a repo in Settings › Connections › Claude Code, commit the two files it adds, and each cloud session there saves its token counts to the repo's headroom-usage branch. Headroom reads them from there and files them under the same projects as your local work. Nothing goes through Headroom's servers.
  • Work an agent does with shell commands, like adding a line to a file, now counts for the folder that file is in, not just work done with its edit tool. Headroom reads your Claude Code history again once to regroup it.
  • Client folders count as projects. In a repo that's a home for several projects, with no package.json or similar at the top, each folder inside a projects folder like Proyectos/, clients/ or apps/ is its own project, and so is any folder with its own CLAUDE.md or AGENTS.md. Headroom reads your history again once to regroup it.
  • Send feedback is now a window in Headroom. Attach a diagnostic report that shows how your project folders are laid out, with every folder name replaced by a letter. Read the whole report before it goes, and get a reference to quote.
  • Good to know: Claude Code cloud sessions don't leave a log on your Mac, so Headroom shows their limits but not which project they were for.
  • Early resets: when Codex, Claude or another tool resets your limits before they're due, Headroom now tells you, however much came back. If it hits several accounts at once, you get one notification.
  • Last call now names one of your other accounts on the same tool with room to spare, like "switch to Work (82% left)".
  • "Runs out" alerts name the day when the reset is days away, and say "about 1 h" instead of "about 60 min".
  • The preview in Settings › Alerts shows every alert, and Send test sends the one you're looking at.
  • Monorepos now split into their projects. When a repo holds several projects, one folder each or a pnpm, Turborepo, Nx or Cargo workspace, each one gets its own row, based on the files Claude Code or Codex actually read or edited. That works even when you start the agent at the repo's root.
  • Folders an agent wandered into, like a website's og folder or a folder inside node_modules, no longer show up as projects of their own. Their work counts for the repo they're in. An app repo with a website folder inside stays one project.
  • Reading large Codex logs takes much less CPU.
  • Headroom reads your Claude Code history again once after this update, and checks Codex history from monorepos, to regroup it.
  • Beta stats: once a day Headroom sends a few anonymous counts, like which version you're on, which tools you've connected and which screens you use, so we can see what works. Never your projects, prompts, token counts or account names.
  • See exactly what's sent, and turn it off, in Settings › General. New installs see the switch at the end of setup.
  • A folder you work in with both Codex and Claude Code is now one project, not two rows with the same name.
  • When a folder is in two of your Codex projects, it goes to the one named after it, instead of whichever came first.
  • Two different folders with the same name now say where each one is, like "in Documents".
  • Choosing one account on the Projects page no longer shows an empty page. Codex doesn't record which account ran a session, so it shows all your Codex work on this Mac and says so.
  • Headroom reads your Claude Code history again once after this update to regroup it.
  • Headroom uses much less CPU while Codex is running. Clearing out history older than 65 days used to look through every saved request each time Codex wrote something; now it goes straight to the old ones.
  • Projects no longer show up under odd names like "agent-ac1bec01d6ea924db". Work Claude Code does in a separate copy of a project, for a background agent or a desktop session, now counts for that project.
  • To fix the names in your history, Headroom reads your Claude Code logs again once after this update. You'll see "Catching up on history…" for a little while.
  • Headroom now checks for updates every hour instead of every four, so new versions reach you sooner. They still download quietly and install the next time Headroom restarts.
  • The dashboard's capacity card now tells you what happens at your pace, in one line: "You'll run out in about 14 hours", or "On track, with about 27% left at the reset".
  • A timeline from now to the reset shows when you'd run out and how long you'd be without it.
  • The two numbers that matter, capacity left and how far ahead or behind you are, sit next to it in plain words.
  • Every limit bar in the menu is now the same thickness. Bars with a pace mark were drawn thicker than bars without one.
  • In setup, the yellow highlights on "How to read a limit" line up exactly with the part they point at.
  • Setup fills its whole window. No more white strip at the bottom.
  • Setup has a few small touches now: each step settles in, ticks land as you finish, and the history bars grow. They're off if you use Reduce Motion.
  • New setup for first-time users: connect your tools, learn how to read a limit, set up the menu bar and alerts. Already set up? Run it any time from Settings.
  • The dashboard now greets you by name, with one line on what needs your attention.
  • After each update, Headroom shows what's new, with the full changelog on headroom.bar. It's also in the … menu any time.
  • Sign in again now only renews that Claude account. If your browser is signed in to a different one, Headroom tells you instead of quietly adding it.
  • If Claude ends Headroom's sign-in, the menu now says so and offers Sign in again, instead of saying it's reconnecting.
  • Signing in again to the same Claude account picks up right where it left off.
  • A dropped connection while Headroom renews your Claude sign-in no longer ends it.
  • Headroom stops checking your limits while your screen is asleep.
  • The even-pace mark on each bar is much easier to see, in light and dark.
  • Account names no longer get cut short when an account has banked resets.
  • If Headroom can't read its saved settings, it now says so and offers to start fresh, instead of asking you to connect a tool.
  • Add more than one account for any tool. In Settings › Connections, open a tool and choose Add account.
  • Codex: sign in with another ChatGPT account, or bring in the accounts you already saved with codex-auth.
  • Claude: add another Claude account next to your first one.
  • OpenRouter and OpenCode Go: add several keys.
  • Cursor, Copilot, Windsurf and Zed: Headroom remembers each account you've signed in with and shows its last reading, greyed out, under the one you're using now.
  • Redesigned menu: every limit shows its pace, a bar with your even-pace mark, what's left and when it resets, all in tidy columns.
  • Tools with several accounts collapse to one line each; open the one you want. Headroom remembers what you opened.
  • Switching 1h, 5h, Today, 7d and 30d now animates the charts smoothly.
  • "Restart to update" now also shows in the dashboard.
  • If Headroom needs to restart twice to reach the newest version, it says "Finishing update…" instead of just restarting again.
  • New: model limits like Claude's Fable weekly limit, with their own line in the menu, a full row in account details and their own alerts.
  • Turn the menu line off in Settings › Menu bar if you prefer.
  • Prices now update on their own: when a new model comes out, Headroom picks up its price within a day and fills in past days, no update needed.
  • A model without a published price yet says so, instead of showing no spend.
  • Corrected prices for several GPT-5 models.
  • The spend and token charts now animate smoothly when you switch accounts.
  • Spend is back for GPT-6 Sol and GPT-6.1 Sol, including the days since you switched to them.
  • Much lighter: Headroom uses about a third of the memory it did, and almost no CPU while you're not looking at it.
  • History is read once and remembered. Restarts and updates no longer re-read everything, and catching up happens quietly.
  • New: "Update ready" in the menu with a Restart button, and it always installs the newest version.
  • Choosing Claude now shows your Claude history from this Mac, not just its limits.
  • The header fits smaller windows, and "+N" opens the accounts that didn't fit.
  • Fixes the "Menu bar" label squeezing into a column when two tools show both limits.
  • Send feedback now includes recent problems (never keys or usage), so fixes come faster.
  • Fixes Headroom quitting as soon as it opened on some Macs.
  • The dashboard header now shows every account with its logo, including Claude. Click one to focus on it, click again for all.
  • New: Account order in Settings › Connections. Drag your accounts into the order you want.
  • The account picker highlights rows as you hover.
  • New: Backup from Claude Code. If Claude sign-in stops working, Headroom shows the limits Claude Code last reported. Turn it on in Settings › Connections › Claude Code.
  • Each GitHub Copilot account now gets its own row, and two accounts can never mix.
  • When a provider asks Headroom to slow down, it waits exactly as long as asked, and only for that provider.
  • A tool that fails to refresh no longer makes Headroom check everything else more often.
  • Connect a tool now opens straight to Connections from the keyboard and VoiceOver too.
  • Connecting a tool can be cancelled, and switching tools mid-way no longer mixes up results.
  • VoiceOver and hover help give exact token totals, not just 1.2M.
  • Clearer help when macOS needs you to approve Open at login.
  • Pace alerts now watch every limit and warn about whichever will run out first.
  • Banked-reset reminders use exactly the reset they're about, and say plainly if it's already gone.
  • Quiet hours now cover every alert.
  • Local history on this Mac is shown separately from your signed-in Claude and OpenCode accounts.
  • OpenCode Go shows an exhausted monthly limit, and saved Windsurf readings are never labelled live.
  • Signing out of Claude can no longer be undone by a refresh that was already running.
  • Send test now shows a fresh notification every time, with sound, and tells you what happened.
  • Never waste a banked reset. In Settings → Connections, each Codex account can remind you before a banked reset expires (with a Use reset button), or use it automatically in its last 2 hours. Off by default.
  • Use one now… spends your soonest-expiring reset when you choose.
  • "Use it before it resets" is now "Don't waste what's left".
  • "Updated…" labels no longer count up second by second: "just now", then minutes.

Alerts, rebuilt

  • Heads-up and Last call for your 5-hour window and your weekly limit, set by dragging markers on a live bar.
  • Warn before I run out: Headroom watches your pace and tells you before a limit runs dry.
  • Quiet hours, per-account alerts, and Snooze or Open Headroom right on the notification.
  • A bell in the menu shows whether alerts are on, with quick snooze.
  • If macOS blocks notifications, Headroom tells you why and takes you to the right setting.
  • If a tool can't be read, a Report this… button sends us the details, so we can fix it fast.
  • Send feedback now includes versions and each tool's status. You see everything before sending, and can switch it off.
  • Settings: the selected tool is easier to spot.
  • Behind the scenes: confirms updates now reach everyone quietly, with no prompts.
  • Updates now download quietly and install the next time Headroom opens. No more update prompts.
  • Headroom checks for updates every few hours during the beta.
  • Headroom's own icon everywhere on macOS 26 and later: Dock, Finder, notifications and system prompts, instead of a small icon on a grey tile.
  • The Headroom disk image shows the Headroom icon when you open it.
  • Behind-the-scenes check that automatic updates reach everyone in the beta.

Welcome to the Headroom beta

  • Every connected tool in the menu bar at once, each with its own logo and reading.
  • Native readers for ChatGPT (Codex), Claude Code, Copilot, Gemini CLI, Cursor and more.
  • When Claude asks Headroom to wait, Headroom says so and checks again on its own.
  • Lower memory and idle CPU.
  • Only one Headroom runs at a time.
  • New: Send feedback… in the … menu.

Give yourself some headroom.

Headroom is in a private beta. Leave your email and we'll send your invite as we let people in.

Your invite and the odd update. No spam, no sharing.

For macOS 15 Sequoia and later · Works with ChatGPT, Claude Code and 9 more