A desktop app that turns a plain-English description (a song, an amp, a pedalboard, a feel) into a complete, valid rig for your HeadRush floor unit, then pushes it to the pedal over your network. Amp and cab pairing, drive stacking, time-based effects, block order and levels, all decided together instead of one knob at a time.
{{ versionLine }} · Free · Runs on your Claude, ChatGPT, Grok or Google Antigravity subscription, or your own API key
“Edge-of-breakup AC30 with a slapback and a swell for the bridge.” Reference a record or a rig, or just say how it should feel.
Blocks are chosen and ordered against your pedal's own model catalog and parameter schema, so what comes back is a rig the hardware actually accepts.
The app finds your unit on the network, checks it, writes the rig as a new preset and reads every parameter back, and you play. Existing presets are never overwritten.
One app on {{ platformsList }}: since 2026.10.2 the Mac runs the same app as {{ otherPlatforms }}, with the same screens, the same wording and the same behaviour, so what you learn on one is true on the others. Only the menus differ, and ⌘ on the Mac where the others use Ctrl. The same model catalog, the same dark design language in two themes, Mint and Black Gold, and the same console, with Build, Rigs, History, Device and Setup down one rail.
Feature tourModelling gear buries its good sounds under menus. The app does that digging for you and hands back something you can plug in and judge with your ears.
Ask for a tone the way you'd ask a bandmate: “Metallica, Master of Puppets, rhythm.” Genre, era, a specific record, a specific amp; the app does the translation into blocks and parameters.
Amp, cab, drive, dynamics, modulation, delay and reverb are picked as a set and drawn as a signal rail, left to right the way signal travels. One card per block, with every parameter the search set and a count of how many of the unit's 14 slots the chain uses. With the unit switched on, each card wears the unit's own picture of that block, fetched from the device as the card is drawn: nothing is bundled into the app and nothing is written to disk. With the unit off, the card keeps its symbol. A doubled amp, cab, IR or clone block is two models in the one slot, and its card shows both sides and which mode the block is in, Stereo or Switch; a push keeps both.
HeadRush's published model attributions and the unit's own parameter schema ship inside the app, so “Mesa Mark IIC+ with the mids scooped” comes back as the pedal's actual 85 M-2 Lead. It cannot reach for a block the hardware does not have, and rigs are re-verified before they're staged. It also plans for the firmware your unit runs, reading it when it reaches the unit and following what that firmware can do in searches, in the palette, in scenes and in the push; out of reach, it uses the last firmware it saw and says so, and having never read one at all it plans for 5.1 and says that too. The search also works to one written build standard, shared by {{ appsPhrase }}: precision where the role pays for it, delay times worked out from the tempo rather than guessed, and a self-check before it answers.
With a web-search-capable backend, the app looks up what gear was actually on the record and tags every claim CONFIRMED, REPORTED or INFERRED, with a source for each. It is forbidden from inventing interviews, quotes or URLs to fill the gaps.
You never move a file yourself. The app finds your unit on the network and writes the rig as a new preset: Push to device, and it saves the rig to your library, checks the unit, builds the rig block by block, saves it there and reads every parameter back. There is nothing to confirm, on any platform, because a push cannot overwrite anything: it lands as “##HRB Name”, or “##HRB Name 2” if that is taken, never in place of anything. Only Save to device, for a rig you opened from the unit, asks first. The HeadRush Remote Editor lives on the Device page inside the app for knob tweaks afterwards, and a found rig even names the real amp it identified and offers NAM captures of it via Tone3000, fetched by the unit itself. The layout and any doubling go first and are checked before the blocks follow, a push that stops part-way says how to recover from where it stopped, and afterwards the app reads the unit's CPU meter and warns you at 100% or more, where the audio can pop. It also names an engaged IR block with no file, which silences the rig.
Seven backends: the Claude CLI, which is the default and runs on the Claude subscription you already have without a key; the Codex CLI on a ChatGPT subscription, the Grok CLI on a Grok one and the Antigravity CLI on a Google Antigravity one, the same way; a Claude API key; a Google key for Gemini; or an OpenAI key. Keys live in {{ keyStores }}. You never create an account, nothing is reported back, and no server of mine sits anywhere in the path.
A scene is the unit's own table: ten scenes, fourteen slots, each block on, off or left alone. The app reads them off a rig, draws the real pad layout for your chassis, and lets you author them where you can see the whole chain. Click a pad and its sheet opens. Close it, turn blocks on or off on the rail, open the pad again and press Capture the rail: the blocks you changed are recorded as that scene, every other block keeping what the scene already said. The same sheet names the pad, colours it, puts its scene back on the rail with Apply to the rail, assigns the picked block to it as a stomp, duplicates the scene to another pad and clears the pad. Scenes save with the rig, every change to them can be undone, and Save to device writes them back the way you laid them out, keeping the pad numbers a Hardware Assign gave them; on firmware 5.2.0 it also undoes the footswitch picks the unit's own Auto Assign Footswitches setting makes during a save. The strip has an Edit and a Live switch: in Edit a click opens the pad's sheet, and in Live a click presses that pad on the unit, so a scene engages, a two-state scene steps between its states and a stomp turns on or off, with the strip showing what the unit reports back. Live runs only while the unit is playing the rig on screen and stops by itself when that stops being true, saying why. On a Switch-doubled rig the sheet also has a Doubling side, No change, A or B, so each scene can pick a side, and a pad can instead be the unit's doubling switch, flipping every Switch-doubled block between A and B; in Live it does that on the unit and reads X2 A or X2 B.
Ask for a song's clean intro, its rhythm and its lead and the search comes back with a rig for each. Press Scene Pick on two or more and they go onto one rig straight away, each on a footswitch pad of its own: three on a Core, on pads 6, 7 and 8, and four on a Prime, on pads 1 to 4. A Flex Prime has only one pad every footswitch mode reaches, so it takes no scene picks. The rig opens as your first pick, its blocks on and the others bypassed, and each pad switches to one of the results. Step 2 lists every pad with the blocks it turns on and off, names anything the merge could not keep, and says plainly when the picks will not fit: a scene can turn a block on or off but cannot change an amp's model, so three different amps need three amp blocks, and a Core has two. Push it like any rig, and the pads come with it.
New in 2026.10.2, on every platform. Every change to the rail can be undone with ⌘Z, or Ctrl+Z on {{ otherPlatforms }}, and redone with ⇧⌘Z, Ctrl+Shift+Z or Ctrl+Y, up to 50 steps, from adding, moving or bypassing a block to applying a capture or setting a scene or a pad; the Edit menu's Undo and Redo do the same. With a block picked, Delete or Backspace removes it and the arrow keys walk along its row. The block palette has a Search models or real gear field: type an amp's nickname or a real pedal's name and only the models that match are left, each naming the gear it is modelled on. With nothing picked, the inspector says what the rig is: its unit and block count, and for a rig from a search the real gear it aims at, how confident the search was, what you asked for and which AI made it. And the Rigs list has a button to read the unit's rig library again, for when you have changed it on the unit itself.
Pick a block on the rail, then press Search beside the Tone3000 mark under the rail: the sheet opens on that block's category, amp, pedal or outboard, with the captures already on the pedal listed a folder at a time and a search for more, run by the unit with your own account. With nothing picked, the capture lands at its conventional position. Since 0.8.26 that door is under the rail, where the block is picked, rather than an item in the nav. A C-Verb can be handed one of the unit's own impulse responses with Choose file, with nothing to download, and the app now reads the User folders all the way down, so files you loaded into a subfolder by hand are found too. A search that wants a reverb names one the unit actually has. How many NAM captures a rig holds follows your unit's firmware: two on 5.2.0, in two blocks or in one doubled block, and one on anything older. The app holds to one until it has read a firmware from the unit at all; after that, a unit it cannot reach is planned for with the last firmware it saw. One C-Verb either way, counted separately, and it takes a .wav, .aif or .opus file. On 5.2 the Pedal Clone block loads the unit's own clone files rather than a NAM capture, so a pedal or outboard capture goes into a NAM block where a drive would sit; Setup lists any rigs in your library that still put a capture in Pedal Clone. A search no longer places an IR block at all, because it cannot see or choose the files on your unit and an IR block with no file is silent. On firmware 5.2 the push now turns on Calibrate Input in a NAM block whose capture the unit can calibrate, unless the rig sets it itself, so the capture's gain sounds as intended; the push window says which blocks it turned on, and says so if the unit did not keep it. Rigs pushed with an earlier version played those captures uncalibrated, so send them again to get it: a rig you opened from the unit goes back over its own preset with Save to device, while a plain push still makes a new preset and leaves the uncalibrated one where it was.
The console puts everything down one rail: Build, Rigs, History, Device and Setup, with a numbered steps panel that tells a first-time user what the app is for. Setup holds the unit, the backend, every API key and the theme; since 0.9.9 it takes the whole window, its sections listed down the left, and the rig on the rail is kept for when you leave it. There was a second, older console until 0.8.28; it is gone, and there is nothing to choose between.
A rig can split into two lanes and join again, and the rail draws it that way: the blocks before the split, the lanes, the join, the blocks after. All five of the unit's path modes render: Straight, the three splits (3-4-3, 5-3-3 and 7-2-3) and Immediate. All five have since been selected on a unit and read back, so none of them carries the question mark the app used to show: each Middle Split is drawn with the slot counts the unit reported, and Immediate Split is the shape the unit actually uses, five and five from the input with four more after the paths rejoin. If you laid a rig out on Immediate Split in an earlier version, the app names the slots it now draws in a different section; the rig on the unit is untouched. A search can answer with two amps on a split, show each one on its path, and list the layouts that fit for you to pick from.
{{ allApps }} apps ask this site, and only this site: Check for Updates reads a small file, latest.json, that the same publish step writes alongside the download page, and when a newer build is listed it offers to open that page, and an Update pill appears in the top bar. That is all it does; nothing is downloaded or replaced behind your back. On the Mac, drag the new build over the old one; on Windows, run the installer over the top{{ linuxUpdateNote }}. Your library and settings stay where they are {{ eitherWay }}. Coming to 2026.10.2 from the old Mac app, the new app sits beside it, and its first launch brings your library across. Releases are named by the year, the month and that month's number: 2026.10.2 is the second of October 2026, and it is newer than 0.9.9. It never points at the source repository, which is private.
Real screens from the Mac build of 2026.10.2, the same app as {{ otherPlatforms }}, in the Mint theme except where a caption says otherwise. The Tone3000, push and fork frames are from the earlier Mac app and their captions say which version; the controls they show work the same way now.
Prime, Core and Flex Prime share the Core's engine, so a rig that fits one fits all three. Point the app at whichever is on your network and every build lands as a valid preset. Footswitch scenes are drawn on the pad layout of the chassis you picked; the pads the unit keeps for itself, previous and next rig in Hybrid mode, are shown and can be renamed, recoloured and cleared, but a scene cannot be recorded onto them.
The renders on the cards above are generated images, not photographs, and this project is not affiliated with or endorsed by HeadRush or inMusic Brands.
| Unit | Rig building | Chain | Transfer |
|---|---|---|---|
| Prime | Full | 14 slots, Core engine | Push over the network (mDNS) |
| Core | Full | 14 slots, the reference | Push over the network (mDNS) |
| Flex Prime | Full | 14 slots, Core engine | Push over the network (mDNS) |
The app reaches the unit by its mDNS hostname, headrushcore.local by default, rather than a raw IP, which goes stale with the DHCP lease. A push always saves a new preset, so nothing on the unit is overwritten.
Start to finish: set up the AI backend the app will use, get the app, build your first rig, push it to the pedal, then pick your theme and keep it current.
The app ships without an AI. When you describe a tone, the app hands your words to a backend you provide and asks it to build the rig, so nothing happens until one of these is set up on this computer. You need exactly one. The four CLIs ride a subscription you may already pay for and need no key; the three keys are billed by their provider per request, except that Google's free tier covers the Gemini Flash models, and none is included in that company's chat subscription. Pick one, set it up, then install the app.
Keys go in Setup → API keys, are encrypted with a key kept in your macOS Keychain, and go only to the provider you chose. Step 04 is where you tell the app which one you set up, and Test backend there makes one real request so you know it works before your first search.
Grab the latest DMG from the download page and drag the app into /Applications. Apple silicon and Intel are both covered by the universal build; it needs macOS 14 Sonoma or newer.
Release builds are Developer ID-signed, notarized and stapled, so Gatekeeper lets it open normally on first launch, even offline. If macOS ever warns you, you're holding a development build rather than a release. macOS may then ask once whether the app can find and connect to devices on your local network: allow it, or the app cannot reach your unit. Coming from the old Mac app? Since 2026.10.2 the Mac runs the same app as {{ otherPlatforms }}, and its first launch brings your library across: your rigs and searches, your unit's address and type, your theme, your AI backend and the API keys you saved. It only reads the old app's data and never changes it; if the Keychain cannot be read just then, locked for instance, it waits and tries again the next time you open it. Two things work differently now: a rig is saved to your library when you open Push, not after every change, and closing the window quits the app.
The first launch asks two things, the backend you set up in step 01 and which HeadRush you have; if your old Mac app's settings came across, it skips them. Change either later in Setup (⌘, opens it) → AI backend. The default, Claude CLI, runs on your existing Claude subscription, and the app finds the CLI wherever it is installed: Claude Desktop's bundled copy, the standard install in ~/.local/bin, or Homebrew's. If you don't have the CLI, paste an Anthropic API key instead (it doubles as the CLI's fallback), or use the Codex CLI on a ChatGPT subscription, the Grok CLI, the Antigravity CLI, a Google key for Gemini, or an OpenAI key. The four CLIs need no key: install each one and sign in once, as in step 01. Keys go in Setup → API keys; they are encrypted with a key kept in your macOS Keychain and sent only to the provider you chose. Test backend makes one real request, so a key that is saved but has no credit on it says so here rather than at your first search, which is the likeliest first stumble: a key is billed separately from the same company's chat subscription, so ChatGPT Plus does not pay for OpenAI API calls. Keys are made at platform.openai.com/api-keys, with credit added under Billing in the same dashboard.
Pick your chassis in Setup → Unit (Prime, Core or Flex Prime) and the app addresses that unit by its own mDNS name. That is the only thing to set, and Check connection names what answered, so you can tell a wrong pick from a quiet network. Leave it as the mDNS name rather than typing a raw IP: the pedal's DHCP lease goes stale, and a hardcoded address makes a working unit look offline. Tone search works with the pedal switched off; only push and Device need it on the network.
Prime headrushprime.local Core headrushcore.local Flex Prime headrushflexprime.local
Describe it, say “Metallica, Master of Puppets, rhythm”, then review the result on the signal rail: every block in signal order, every parameter the search set, and how many of the unit's 14 slots it uses. Type a name in the inspector's Name field on the right; it applies as you type. Pick something you'll recognise on a dark stage.
Press Push to device, step 3 of Build: the sheet saves the rig to your library, checks the unit, builds the rig on it block by block, saves it and reads every parameter back, then takes you to Device. There is nothing to confirm. It always saves a new preset, “##HRB Name”, or “##HRB Name 2” if that is taken, and nothing on the unit is overwritten. Knob tweaks afterwards happen in Device, which is the HeadRush Remote Editor running inside the app.
The console puts Build, Rigs, History, Device and Setup down one rail. Theme, Mint or Black Gold, is picked from the right-hand end of the footswitch strip, or in Setup. The screenshots on this site are in Mint except where a caption says otherwise.
HeadRush Rig Builder → Check for Updates… asks this site for its latest.json and, when a newer Mac build is listed, offers to open the download page; an Update pill also appears in the top bar. That is all it does: nothing is downloaded or replaced by the app. Drag the new build over the old one and your library and settings stay where they are, in ~/Library/Application Support/HeadRush Rig Builder; File → Open Data Folder takes you there.
A GUI app launched from the Dock doesn't inherit your shell's PATH, which used to hide a Claude CLI installed in ~/.local/bin. The app looks in the install locations itself: Claude Desktop's bundled CLI, the standard install in ~/.local/bin, then Homebrew's /opt/homebrew/bin, taking the first that actually runs. What still hides it is a shell alias or function named claude, which no spawned process can see. Put the CLI's directory on the PATH the app itself sees, or paste an Anthropic API key in Setup → API keys, which needs no CLI at all.
On an API backend, usually the key: check it's pasted whole, still active, and has credit on the provider side. Under Claude CLI, a saved key is the automatic fallback when the CLI fails.
Same network as the Mac, and reached by hostname. A second unit of the same model shows up with a suffixed mDNS name, which goes in A different hostname on the Device page. Guest and isolated Wi-Fi networks often block mDNS entirely. If you answered Don't Allow when macOS asked about the local network, turn the app back on in System Settings → Privacy & Security → Local Network.
File → Open Data Folder, then logs/server.log. That's the file to attach to a bug report.
The same shape as the Mac guide, with the paths and prompts Windows actually shows you. You don't need any developer tools.
The app ships without an AI. When you describe a tone, the app hands your words to a backend you provide and asks it to build the rig, so nothing happens until one of these is set up on this computer. You need exactly one. The four CLIs ride a subscription you may already pay for and need no key; the three keys are billed by their provider per request, except that Google's free tier covers the Gemini Flash models, and none is included in that company's chat subscription. Pick one, set it up, then install the app.
Keys are stored encrypted with Windows DPAPI and go only to the provider you chose. The first launch asks which backend you set up (step 04), and Test backend in Setup checks it without running a full search.
Take the installer from the download page. It's 64-bit, for Windows 10 or 11. There's also a portable exe that runs without installing anything.
The build isn't code-signed yet, so Windows shows “Windows protected your PC”. Click More info → Run anyway. If your browser blocked the download, keep it from the downloads list first. The installer needs no admin password and puts a shortcut in the Start menu.
First launch asks two things: how tone search should think, which is the backend you set up in step 01 (your Claude subscription via the Claude Code CLI, which is the default, your ChatGPT subscription via the Codex CLI, a Claude API key, your Grok subscription via the Grok CLI, your Google Antigravity subscription via the Antigravity CLI, a Google key for Gemini, or an OpenAI key), and which HeadRush you have. Both are skippable and changeable later in Setup, where Test backend checks a backend without running a full search, and that check is worth running: an API key is billed separately from the same company's chat subscription, so ChatGPT Plus does not pay for OpenAI API calls and a new key answers nothing until credit is added at platform.openai.com. A saved key is stored encrypted with Windows DPAPI, never in plain text.
Type what you're after, say “Metallica, Master of Puppets, rhythm”, watch the progress stream, then review the staged result on the signal rail: every block in order, every parameter set, and how many of the unit's 14 slots it uses. Give it a short, stage-readable name.
Push to device, step 3 of Build: the app saves the rig to your library, checks the unit and pushes, with nothing to confirm, and it lands as a new preset. It is saved as “##HRB Name”, or “##HRB Name 2” if that is taken, and existing presets are never overwritten. Knob tweaks afterwards happen in Device, the HeadRush Remote Editor running inside the app.
File → Check for Updates asks this site for its latest.json and opens the download page when something newer is listed; an “Update” pill also appears in the top bar. It only ever talks to this site. Installing over the top keeps your library and settings, and the app never updates itself behind your back. Everything the app saves lives here (File → Open Data Folder takes you there):
%APPDATA%\WinHeadrushRigBuilder
Windows draws the same console as the Mac: Build, Rigs, History, Device and Setup down one rail. Theme, Mint or Black Gold, is picked from the right-hand end of the footswitch strip, or in Setup. The screenshots on this site are from the Mac build; the Windows one draws the same console.
Unsigned installers trip heuristics. Restore it and add an exclusion for the install folder.
Reach it by hostname, headrushcore.local, rather than by IP: the unit answers over mDNS, and its IPv4 lease can be stale while the hostname still works. Both machines need to be on the same network, and guest or isolated Wi-Fi usually blocks mDNS.
File → Open Data Folder, then logs\server.log. That's the file to attach to a bug report.
A Linux build, an AppImage and a .deb for x86_64, is being tested. This guide and its download appear the day the first release is public. Until then the Mac and Windows guides are the ones to read.
The same shape as the Mac and Windows guides, with the commands and notices Linux actually shows you. An AppImage or a .deb, for x86_64; no developer tools needed.
The app ships without an AI. When you describe a tone, the app hands your words to a backend you provide and asks it to build the rig, so nothing happens until one of these is set up on this computer. You need exactly one. The four CLIs ride a subscription you may already pay for and need no key; the three keys are billed by their provider per request, except that Google's free tier covers the Gemini Flash models, and none is included in that company's chat subscription. Pick one, set it up, then install the app.
Keys are stored through your desktop keyring, GNOME Keyring or KWallet, and go only to the provider you chose. If no keyring is running, the app tells you at first launch that a saved key would not be encrypted; the four CLIs keep their own sign-ins and are not affected. The first launch asks which backend you set up (step 04), and Test backend in Setup checks it without running a full search.
Take either from the download page. Both are x86_64 and need glibc 2.33 or newer: Ubuntu 22.04 and later, Debian 12 and later, Fedora 34 and later, but not Ubuntu 20.04 or Debian 11. On Ubuntu and Debian take the .deb: it installs Chromium's sandbox helper, puts the app in your menu, and asks for the mDNS resolver the unit's hostname needs. Anywhere else, the AppImage is one file that runs without installing anything.
The .deb installs with sudo apt install ./headrush-rig-builder_<version>_amd64.deb; then start HeadRush Rig Builder from your menu. The AppImage needs to be made executable once, then it runs from wherever you keep it, by double-click or from a terminal:
chmod +x HeadRushRigBuilder-<version>-x86_64.AppImage ./HeadRushRigBuilder-<version>-x86_64.AppImage
It needs no libfuse2. On Ubuntu 23.10 and later, AppArmor refuses the unprivileged user namespaces Chromium's sandbox asks for, and the AppImage's launcher then runs the app without that sandbox; the .deb keeps it, which is why it is the pick there. Neither build is signed, so there is no signature to check and nothing to click past; take them from this site only. The first launch may show one or two notices, about the keyring and about .local names; each says what to install, and each is shown once.
First launch asks two things: how tone search should think, which is the backend you set up in step 01 (your Claude subscription via the Claude Code CLI, which is the default, your ChatGPT subscription via the Codex CLI, a Claude API key, your Grok subscription via the Grok CLI, your Google Antigravity subscription via the Antigravity CLI, a Google key for Gemini, or an OpenAI key), and which HeadRush you have. Both are skippable and changeable later in Setup, where Test backend checks a backend without running a full search, and that check is worth running: an API key is billed separately from the same company's chat subscription, so ChatGPT Plus does not pay for OpenAI API calls and a new key answers nothing until credit is added at platform.openai.com. A saved key goes into your desktop keyring, never into a plain file, unless the app has told you there is no keyring to use.
Type what you're after, say “Metallica, Master of Puppets, rhythm”, watch the progress stream, then review the staged result on the signal rail: every block in order, every parameter set, and how many of the unit's 14 slots it uses. Give it a short, stage-readable name.
Push to device, step 3 of Build: the app saves the rig to your library, checks the unit and pushes, with nothing to confirm, and it lands as a new preset. It is saved as “##HRB Name”, or “##HRB Name 2” if that is taken, and existing presets are never overwritten. Knob tweaks afterwards happen in Device, the HeadRush Remote Editor running inside the app.
File → Check for Updates asks this site for its latest.json and opens the download page when something newer is listed; an “Update” pill also appears in the top bar. It only ever talks to this site, and it never replaces anything itself. With the AppImage, download the new one and run it in place of the old; with the .deb, install the new package over the top. Your library and settings stay put either way, here (File → Open Data Folder takes you there):
~/.config/headrush-rig-builder
Linux draws the same console as the Mac and Windows: Build, Rigs, History, Device and Setup down one rail. Theme, Mint or Black Gold, is picked from the right-hand end of the footswitch strip, or in Setup. The screenshots on this site are from the Mac build; the Linux one draws the same console.
The unit announces itself as headrushcore.local over mDNS, and Linux only resolves such names through a module. Install avahi-daemon and libnss-mdns (sudo apt install avahi-daemon libnss-mdns on Debian and Ubuntu, sudo dnf install avahi nss-mdns on Fedora), check that the hosts: line of /etc/nsswitch.conf includes mdns4_minimal, and restart the app. Ubuntu desktop ships both already.
Reach it by hostname, headrushcore.local, rather than by IP: the unit answers over mDNS, and its IPv4 lease can be stale while the hostname still works. Both machines need to be on the same network, and guest or isolated Wi-Fi usually blocks mDNS.
There is no desktop keyring for the app to use, so a key you save would be readable by anyone who can read your files. Install and unlock GNOME Keyring (gnome-keyring) or KWallet, restart the app, and enter the key again. The Claude, Codex and Grok CLIs keep their own sign-ins and are not affected.
Check it is executable (step 03). If the distro is older than glibc 2.33, Ubuntu 20.04 or Debian 11 for instance, the app's engine cannot start at all; that needs a newer release of the distro, not a newer app.
File → Open Data Folder, then logs/server.log. That's the file to attach to a bug report.
Free, built from source you can read. {{ versionLine }}.
Universal build for Apple silicon and Intel, macOS 14 or newer. Signed, notarized and stapled, so it opens without Gatekeeper warnings; macOS may ask once for local network access. Coming from the old Mac app, the first launch brings your library, settings and keys across.
64-bit for Windows 10 and 11. No admin rights needed; a portable exe is there too if you'd rather not install.
x86_64, as an AppImage or a .deb, on glibc 2.33 or newer. On Ubuntu 23.10 and later the AppImage runs without Chromium's sandbox, so prefer the .deb on Ubuntu and Debian. Unsigned, like the Windows build.
The Mac build is notarized and opens without Gatekeeper warnings. The Windows build isn't code-signed yet, so SmartScreen warns on first run; the Windows guide covers the two clicks past it.{{ linuxSigningNote }} The downloads and the notes above come from the public releases repository; the source repository stays private.
Every button in the console, what it does, and where it is. Since 2026.10.2 it is one app on {{ platformsList }}, so the screens, taken from the Mac build of 2026.10.2 in the Mint theme unless a caption says otherwise, are what you see on every platform; only the menus differ, and Ctrl stands in for ⌘ on {{ otherPlatforms }}.
Build is the panel the app opens on. Type into Describe the tone, a record, a song, a player or the gear itself, and press Build this rig. The button is off until there is text in the box, and reads Searching… while the search runs. Anything it could not do is written in red under the button rather than hidden.
The three numbered steps under it, Describe it, Pick a rig and Push to device, are the whole job; the step you are on is lit and the others are dimmed. While the search runs the panel reads Searching… 21 s beside a pulsing dot and the button reads Searching…; when it ends it says how it ended and how long it took, Done in 34 s: 3 rigs to choose from., no rig found, or Stopped after 12 s. A rig opened from Rigs or History counts step 1 as done, because you did not need to describe it.
An ambiguous ask comes back as a short list, 3 rigs to choose from, each with a line on what sets it apart. Press Pick this rig and it is on the rail at once, ready to edit. Pick is a toggle, so pressing it again puts the rail back to what it was before the results came in. Scene Pick is a second toggle on each result: pick two or more and they are combined on one rig as scenes straight away, three on a Core, on pads 6, 7 and 8, and four on a Prime, on pads 1 to 4, while a Flex Prime has one pad for it and so takes no scene picks. Each pick gets a pad of its own, and the rig opens as the first pick with the others' blocks bypassed. Step 2 lists each scene's pad with the blocks it turns on and off, anything the merge did not keep, and why a pick would not fit, such as a third amp on a unit with two amp blocks. A picked rig and Scene Picks are two different rigs, so Next waits until you clear one of them. A result with two amps asks for its split layout right under it. Next takes you to step 3 once your picks make exactly one rig, and says why when they do not; step 2 stays usable after it, so you can still change your mind. If a pick would replace a rig on the rail that has changes you would lose, it asks first and names the rig; trying results one after another without editing them does not ask. Cancel and start over, on steps 2 and 3, empties the rail and the tone box and takes you back to step 1. The search stays in History either way.
While a search runs, or a rig is being read off the unit, the rail cannot be edited. A new search, a result opened from History, or a saved rig or a rig on the unit that you open all ask the same way before replacing a rig on the rail you would lose changes to.
Each card is one block: its type across the top, the unit's own picture of it when the unit is on, and the model's name. Click a card to pick it; the ring is the selection. Along its foot: Bypass (or Engage, when it is off; a bypassed card dims), Swap…, which changes the model and keeps the slot, and the × that removes it. Drag a card to reorder. The spacebar bypasses the picked block whenever you are not typing. Delete or Backspace removes it, and ← and → move the pick along its row; with nothing picked, → picks the first block and ← the last. ⌘Z (Ctrl+Z on {{ otherPlatforms }}) undoes a change to the rail, up to 50 steps, and ⇧⌘Z (Ctrl+Shift+Z or Ctrl+Y) redoes it; opening a rig or a search's new rig starts the history fresh. None of these keys act while you are typing, with a sheet open, or on Setup.
Add a block, under the rail, opens the palette: the unit's categories and every block in them, each naming the real gear it stands for, with a Search models or real gear field at the top. Type an amp's nickname or a real pedal's name and only the models that match are left. The count, N of 14 blocks, is in the top bar. An empty cell's dashed + adds into that exact position and lane. Anything the hardware would refuse, a fifteenth block, a second NAM capture, is named on the rail rather than silently dropped.
Above the rail, the path modes: Straight, 3-4-3, 5-3-3, 7-2-3 and Immediate. Pick one and the rail redraws as that fork. Each of these has been selected on a unit and read back, so the rail draws the slot counts the unit reported rather than an inferred pairing, and the question mark earlier versions showed beside the splits is gone. Immediate splits straight from the input, five and five, with four more slots after the paths rejoin.
The inspector on the right starts with Name: type, and the rig is renamed as you type; that is the name the unit will show, prefixed ##HRB. With nothing picked, under it is what the rig is: its Target unit, its Blocks, and for a rig that came from a search the Real gear it aims at, the search's Confidence, the Query you asked and which AI it was Searched with, with the search's Mapping notes under a disclosure. Pick a block and you see its model, the real gear it models, Bypass and Swap…, and the values the rig sets on it, as read-only dials; for the blocks whose sound is a file, a Choose file… button (Change file… once one is loaded). Knobs are turned on the unit, on the Device page.
The bar under the rail with the Tone3000 mark is the door. Pick the block you want the capture to replace, then press Search: the sheet opens on that block's category, amp, pedal or outboard, lists the captures already on the unit a folder at a time, and searches Tone3000 for more. The unit does the searching and downloading with your own account. With nothing picked the capture lands at its conventional position, and the note beside the button says so.
A C-Verb takes one of the unit's own impulse responses instead, from Choose file… in the inspector; nothing is downloaded. And when a search names real hardware that the unit holds a folder for, a sentence and a button per matching folder appear under the rail: the reverb offer. Press one and the browser opens on that folder; nothing loads until you pick a file. How many NAM captures a rig holds follows your unit's firmware: two on 5.2.0, either in two blocks or in one doubled block, and one on anything older. The app holds to one until it has read a firmware from the unit at all; after that, a unit it cannot reach is planned for with the last firmware it saw. One C-Verb either way, counted separately, and it takes a .wav, .aif or .opus file. On firmware 5.2 a pedal or outboard capture goes into a NAM block rather than Pedal Clone, which loads the unit's own clone files there.
The strip along the bottom is your chassis's pads, headed Footswitch scenes. In Edit, click a pad and its sheet opens. To capture a scene, close the sheet, turn blocks on or off on the rail, open the pad again and press Capture the rail: it adds what changed since you first opened this pad, or since your last capture on it, to what the pad already holds, every other block keeping what the scene already said. The sheet also has the pad's Name, the label it will wear on the unit after a push, its Colour from the unit's own numbered colours, Apply to the rail, which puts the pad's scene on the rail and can be undone, Assign with the picked block's name, which makes the pad that block's stomp, Clear assignment, Duplicate, and Clear the pad. It lists every block the scene turns on or off, and says what a long press on that pad does on your chassis in each footswitch mode.
Only the pads the unit can press in every footswitch mode are yours to author. A Prime or Flex Prime shows an EXP pedal label for its built-in pedal, which is assigned on the unit. Scenes save with the rig and Push writes them, keeping the pad numbers a Hardware Assign gave them. The header carries an Edit and a Live switch. In Live, a click presses the pad on the unit instead: a scene engages, a two-state scene steps, a stomp goes on or off, and the strip shows what the unit reports rather than what you set. Live needs the unit to be playing the rig on screen, and it ends by itself when that stops being true and tells you why, as it does when you open Setup, which hides the strip; a pad whose press the app has not measured says so rather than guessing. On a Switch-doubled rig the sheet also carries a Doubling side, No change, A or B, and the pad shows the side its scene picks, and This pad is: Scene / Doubling switch: as the doubling switch, the pad flips every Switch-doubled block between side A and side B. It starts labelled A<X2>B, takes the unit's own colour or one you pick, and Clear the pad removes it. After a push the pad shows the label you gave it; once the rig is reloaded on the unit it shows the unit's own label instead, keeping its switch, its latch mode and its colour, and the pad's sheet says so. A pad is one or the other, never both, and in Live pressing it switches the side on the unit.
Step 3 of Build. Since 0.9.8 there is no Push button in the top bar: to push a rig you changed on another panel, go back to Build. For a rig you opened from the unit the button reads Save to device, and that writes back to the preset the rig came from, after asking. The rest of this step is the push: a rig the app built. It opens a sheet that saves the rig to your library and checks the unit (Checking the device…); if something is wrong it says Not ready to push with each reason on its own line, and offers Check and push. If the unit is ready it goes straight on, with nothing to confirm: a new empty rig is started on the unit, the blocks are filled in one at a time, it is saved as a new preset and every parameter is read back. It is saved as ##HRB Name, or ##HRB Name 2 if that is taken, and a push never writes over a preset that is already there. Save to device is the one write that goes back over a preset that is already there: the rig you opened names which preset that is, and saving is what replaces it, after asking. What it writes is the rig as the app has it. The unit does not report a knob turned or a block bypassed on the hardware itself as an unsaved change, so the app cannot know about one and will not warn you: anything changed on the unit, or in the Remote Editor, since the rig was read is written over. Save to device on firmware 5.2.0 meets one more thing: with the unit's Auto Assign Footswitches setting on, the unit puts each block the save rebuilds onto the next free footswitch by itself, from switch 6 on a Core, where that was measured; no guide says where it starts on a Prime or a Flex Prime, and the save checks every switch either way. Since 0.9.9 Save to device waits for those picks, sets each switch your rig had nothing on back to unassigned (never a scene switch), and then puts the rig's own scenes and pads back. If a pick is still on a switch when it saves, its window names the switch and the block and says the rig was saved with it; if the app could not read the footswitches at that moment it says that instead, and Hardware Assign on the unit is where to check. The setting is left as you set it, and with it off Save to device works as before. A push never asks first, on any platform, because it never writes over anything; Save to device asks Save over “Name” on your device?, then Yes, save over it.
When it lands the sheet says “Name” is on your device and how many parameters were written and read back correctly, and the window goes to Device, because that is where the next thing you do with it happens. A read-back that disagrees stays on screen with the slots that differ, so a half-written rig is never waved through.
Device in the nav opens the unit's own Remote Editor, live, inside the window; the bar above it says the one thing to know, The device's Remote Editor. Knobs here change what is loaded on the hardware. Turn a knob here and the hardware moves. Builder takes you back to the console. Beside it, Sync from device re-reads the unit's block catalog, Read rigs from device reads its rig list, the firmware line names what the unit runs, A different hostname reaches a second unit of the same model, and Open in its own window gives the editor a window of its own; View → Reload reloads the page when the editor itself misbehaves. Device only exists with the unit on the network, as do captures, the push, the unit's rigs and the live pads; Build, the library, History and Setup work with it off.
Since 0.9.8 this panel is the unit, and only the unit. The app's own library moved to History, under Saved rigs. What is left here is the unit's own list, read live off the hardware in the unit's own order, headed On the unit; since 2026.10.2 the ↻ beside the heading reads the list again, for when you have changed it on the unit itself. A rig that is already in your library opens from there, and the rig the unit is playing is read without loading it; any other rig asks first, because Load and read changes what the unit is playing. Rigs read off the unit save and push again since 0.9.8. Many of them, older imports above all, used to be refused: a control an older version stored as a number or a switch as 0 or 1 is now read the way the unit means it, a time or rate read while its block was synced to tempo is left to the unit and the push says so rather than refusing an impossible value, ReValver Cab's Proximity is no longer refused for running backwards, and a rig with no amp block at all, such as the factory vocal, bass and synth presets, goes through with a note. A rig read on firmware 5.1 that uses the NAM block's Normalize setting is still refused on a 5.2 unit, with the block named. Listing changes nothing, and nothing opens while a search is running or a rig is being read off the unit.
History is one of the five panels down the rail, and since 0.9.8 it holds the searches and the library both; there is no separate window to open. Searches, at the top, lists every search you have run. The ones that found rigs carry every option each produced - click one to open it on the rail, and an option you kept says saved as and the name. Since 0.9.1 the ones that found nothing are kept too, the search's own words as the heading and under it the reason it gave for declining, in full; a refusal is usually saying something about the question. Saved rigs is the app's own library, which lives here since 0.9.8: every rig saved in the app with its block count, an Export beside it and an × to remove it, and Import file… under the list. Click the rig itself to open it: it goes on the rail ready to edit and the console moves to Build, and the search stays in History with the other results still on it. The × on a search takes its options and leaves any rig saved from it; the × on a rig with blocks on it asks first, and a deleted rig cannot be brought back.
A rig can leave the app as a file and come back as a new rig, on any platform the app runs on. Export beside a saved rig writes one; Export beside the rig the unit is playing, over in Rigs, writes that rig as it plays now, saved or not, and writes nothing to the unit. If the unit changes rig while it is being read, the export is refused rather than mixing two rigs. Import file… always adds a rig: it never replaces one and never touches the unit, and if a rig of that name is already there the new one is added beside it and the app says so. A file does not carry the unit's own .rig file, or your capture, clone, IR and reverb files, which it names but does not include, so the unit you push to needs the same files; a scene's block presets are not pushed from a file, and the import says which scenes name one. A file never contains your unit's network name or address.
Since 0.9.9 Setup takes the whole window: opening it replaces the rail, the inspector and the footswitch strip with one settings page, and the column on the left lists its sections, Unit, AI backend, API keys, Theme, Knob curves, NAM captures in Pedal Clone and Run the setup steps again; press one to jump to it. Your rig, its selection and its scenes are kept, and are back as they were when you leave Setup. The strip is hidden while Setup is open, so a Live session on it ends when you open Setup. Unit: the Chassis, Prime, Core or Flex Prime, the hostname the app is reaching, headrushcore.local for a Core, and Check connection. AI backend: the seven tone-search backends, Test backend, and whether the picked backend's key is saved. API keys: the Claude, Google and OpenAI keys, all three whatever backend you use. Each says Saved or Not set, has a field for a new key, and has Remove saved key when one is saved, each remove button naming its key; Save keys saves only the keys you typed, with the backend you last saved, so a backend you clicked but did not save is never saved along with a key. Before 0.9.9 a key field showed only under the API backend you had picked, so on a CLI backend, the default, there was nowhere to see or change a key. A saved key stays on the computer running the app, is never shown again once saved, and is sent only to its own service when that backend runs a search. Theme: Mint or Black Gold, which since 0.9.8 is also at the bottom right of every other panel, at the end of the footswitch strip; a pick applies at once either way. Knob curves: if you imported rigs from the unit with an older version, this is where you settle the curves a later version reads correctly, which the app asks for before it will push such a rig. It names the curves it is settling, asks when you installed the version that corrected them, and previews each repair before you commit it; none should be run twice, and a library with nothing read off the unit has nothing to settle. NAM captures in Pedal Clone lists any rigs that still put a NAM capture in a Pedal Clone block, which firmware 5.2 will not load, and Run the setup steps again brings back the first-launch steps, the backend and then the unit.
There is no separate Settings window: on the Mac, ⌘, and HeadRush Rig Builder → Settings… open Setup. A second unit of the same model goes in A different hostname on the Device page, and Sync from device there re-reads the unit's block catalog. Saved keys are kept in {{ keyStoresInline }}.
| Space | Bypass or engage the picked block, unless you are typing. On a focused Live pad, press the pad instead. |
| Escape | Close a sheet, unless it is writing to the unit. |
| Return | On a focused Live pad, press it; Space works too. |
| ⌘Z | Undo the last change to the rail, a scene or a pad included, up to 50 steps. |
| ⇧⌘Z | Redo what ⌘Z undid; Ctrl+Y also works on {{ otherPlatforms }}. |
| Delete | Remove the picked block from the rail; Backspace too. |
| ← → | Pick the block before or after the one picked now, along its row; with nothing picked, → picks the first block and ← the last. |
| ⌘, | Open Setup (Mac). |
| Menus | Mac: HeadRush Rig Builder → Check for Updates… and Settings…, and File → Open Data Folder. {{ otherPlatforms }}: File → Check for Updates… and File → Open Data Folder for the library, settings and logs. Edit → Undo and Redo on every platform. |
Ctrl stands in for ⌘ on {{ otherPlatforms }}. None of these keys act while you are typing, with a sheet open, or on Setup.
The pedal's built-in Remote Editor is a web page served by the unit, and it works by calling an API the unit exposes on your network. The app talks to that same API directly, by the unit's mDNS name (headrushcore.local, headrushprime.local or headrushflexprime.local), with no account and no server of mine in between.
A push never touches an existing preset. The app starts a new empty rig on the unit, fills in the blocks one at a time, saves it as a new preset named ##HRB <your rig name>, and if that name is already taken adds a 2, 3, 4 and so on. Then it reads every parameter back and tells you if anything did not take.
The same API is how it reads the captures and impulse responses already on your unit, fetches the unit's own block pictures for the rail, reads your footswitch assignments, and lists the rigs on the unit. Tone3000 downloads run on the unit itself, with your own Tone3000 account. The Device tab is the real Remote Editor page loaded inside the app, so knobs there change the unit immediately, same as in a browser.
For tone search, and for Check for Updates. The search goes to the AI backend you chose, and with web search on it verifies gear against real sources; the update check reads one small file from this site and nothing else. Pushing to the pedal is local-network only, and your library works offline.
The app is free. On the default backend, searches run through the Claude CLI on the Claude subscription you already have, so there is no key to buy and no extra bill. On an API-key backend you pay your provider per request instead. An API key is not the same thing as that company's chat subscription. ChatGPT Plus does not include OpenAI API access, and a Claude or Gemini plan does not cover their APIs either: an API key is billed separately and a new one needs its own credit before it will answer anything. The four CLI backends, Claude, Codex, Grok and Antigravity, are the ones that ride a subscription you are already paying for.
No. There's no account, and no server of mine sits in the path. Search queries go to the backend you picked, and what that provider does with them is between you and them: the Antigravity CLI, for one, counts each search against your Antigravity quota and records it in Antigravity's own conversation history. Rigs, settings and keys stay on your machine, with keys in {{ keyStores }}.
Not unless you opened the rig from the unit and saved it back, which writes to the preset it came from and asks you first. One caveat to that ask: it covers the preset, not the knobs. The unit does not report a knob turned or a block bypassed on the hardware as an unsaved change, so anything changed on the unit, or in the Remote Editor, since the rig was read is written over without a warning. Otherwise no: a push always saves a new preset, named ##HRB <your rig name>, and if that name is already on the unit the new one gets a 2, 3, 4 on the end. A push never touches a preset that is already there, and the app reads the new one back to prove it landed. The first question above has the whole mechanism.
The gear and broad settings for a well-known tone are well documented, and that part holds up. With web search on, every claim is tagged as confirmed, reported or inferred, with a source. The exact knob positions on anyone's amp are not documented anywhere, and your guitar, pickups and speakers are half the sound. Treat the result as somewhere sensible to start, then use your ears.
Prime, Core and Flex Prime, which share the Core's engine, are the only units the app targets. Nothing is tuned or tested for anything else.
The pedal's own API is unauthenticated, so anyone on the same network can read and write its state. That's the hardware, not this app. Keep the unit off shared and guest networks.