Talk to your rig.
It builds the tone.

Build a whole rig from one sentence

• A complete multi-scene preset from a single prompt: the signal chain wired, an amp voiced per scene role, a cab chosen, drives and effects placed, and the scenes named on your footswitches.

  "Build a modern metalcore preset in drop C: a big ambient clean, a tight chug rhythm, a lead that sings over the mix, and a heavier breakdown."

• One tap on + Create drops a blank rig on a free slot (or clears the loaded buffer on a full unit) so you always have a clean canvas to build on.

• Open-ended ask? It asks a couple of quick questions first (era, how many scenes, how heavy). Specific ask? It just builds.
• The eight-scene grid fills in live as you talk it through, so you catch a gap ("nothing in scene 8 yet") before it is built.
• A tone rulebook rides on every build: cleans come out loud, leads out-saturate the rhythm, no scene ships silent. That balance was dialed in by ear on real hardware, so a cold build lands close on the first try.

Build a tone from a video you do not have time to watch

• Paste a YouTube link to somebody's rig rundown and get a plan against your own FM9 without watching it. When the video has captions (nearly all do, whether the creator published them or YouTube auto-generated them) it reads the whole transcript and pulls out only the parts that matter in about twenty seconds: the intro, the sponsor read, the merch plug and every tangent are thrown out, and a two-hour walkthrough becomes a compact tone spec.

  "build the rhythm tone from this rig rundown: youtube.com/watch?v=..."

• No captions? It transcribes the audio locally, on your machine, in a few minutes (up to about ninety minutes of runtime; past that it asks you to paste the part that matters).

• A forum thread or a pasted chat about someone's tone works the same way: drop in a URL or the text of the conversation and it does the same read-and-extract.
• It marks which numbers the source actually stated and which it chose, so a guess never wears the source's authority. Ballpark to review and tweak, not a clone, and a build from someone else's video never stores to a slot on its own.

Voice it in plain language

"a Klon into a JCM800 with a greenback 4x12"
"tighten the gate for drop C and bump the presence slightly"
"make scene 1 a dry crunch rhythm and keep the wets in scene 2"

• Pick a cabinet by name, not by scrolling thousands of IRs: "put a 4x12 V30 on the rhythm channel."
• Per-block manual controls with -/+ fine-tune on every slider, a graphic EQ that looks like one, and 331 amps searchable by the real gear they model.

Borrow tone from the tones you already own

• Lift an effect off another preset, aligned scene-for-scene, not just channel for channel: "make my delay like BT Steve Lukather's."
• Compose a preset from parts of several artists' tones:

  "clone BT Marco Sfogli into slot 21, take the gain stage from it, the delay from the Andy Timmons tone, and the reverb from the Periphery one." Every source is read wire-for-wire first, then the new preset is assembled and stored once. A better path to a pro tone than guessing parameters from scratch.

• Reorder blocks into the right signal-chain order (drive before the amp, delay before reverb), the path proven intact.

Two expression pedals, by name

"put global volume on Pedal 1 and wah on Pedal 2"
"delay mix on Pedal 2 so I can swell into the chorus"

• Bind either onboard pedal to any continuous parameter, and unbind it the same way, all by typing.

Fast, and honest about it

• A build lays its whole starter chain in one pass instead of splicing blocks in one at a time, and verifies a block's parameters in a single read instead of one-at-a-time. The slow, fragile part is gone.
• A health scan after every build flags a dead or duplicated scene the moment it lands. Undo and A/B the FM9 itself does not have.
• Install artist presets and cabs from Fractal's Gift of Tone. The catalog shows only what your unit can take and says in one line when a pack needs newer firmware; the zip is fetched from fractalaudio.com by your own app at click time, checked against the catalogued sha256 before it is opened, and cached by that hash. Nothing of Fractal's is mirrored here. "Sound like Devin Townsend" checks the catalog first and offers the real pack; a name with two matches asks which; a name with no pack says so plainly and the build that follows is called an interpretation, never their preset. The ARTISTS section in the Storage drawer installs a pack in one click: cabs into free user-cab slots (moved, and the preset repointed, when the pack's own slots are taken), the preset into a free store slot, and one line saying where everything landed. Paste an Axe-Change preset link and ToneCommand reads that one page, says what it is, checks it is for your unit, and installs it the guarded way. An installed pack then works as a reference: "difference between scene 1 and Devin's" reads the pack's preset off the unit and comes back to your own with your unsaved edits put back (and says so if a block placed unsaved could not be), and the planner's reference lists each installed pack's amp, cab and drives, tagged with the artist and year, as facts read from the file, never as the artist's words.

Measures, with a named baseline

• The FM9 is a USB audio interface, so its output can be recorded and measured instead of assumed. tools/measure.py <capture.wav> (or /api/measure) answers band ratios over six regions, BS.1770 loudness, stereo width and dynamics, after an acquisition check (silence, clipping, dropout) that can make the whole measurement invalid. A number is evidence next to the baseline it was measured against ("2.4 dB more bite than your saved rhythm reference"), never a verdict on its own; the only pass/fail rules are the relative scene-balance ones the rulebook states, in config/sound_policy.json. SOUND CHECK, under the health scan, records each scene through the USB path, measures it, and proposes the exact OUTPUT_SCENEn move for anything the rules catch; the proposal goes through the same Confirm as every plan, then the next SOUND CHECK verifies it, three rounds at most, and says what is left. MATCH REFERENCE compares the capture to a clip of your own (a song, a record) and proposes the amp knob and direction for each band that differs, one step at a time. it, three rounds at most, and says what is left.
• Sign in to your own TONE3000 account from Settings (their documented OAuth flow) and a recipe's capture is fetched under your entitlement; anything your account cannot reach is a link, never a fetch through us.

Never sends behind your back

• Read-back verification on every write, and a signal-path check so a preset is never silently dead.
• A "what will change" blast-radius preview before anything is sent, including every other scene a channel-level change would move.
• Gig lock: mid-set, only scene changes reach the amp.
• Store only to slots you have marked disposable, and only after you confirm.
• Bring your own AI: ChatGPT, Gemini, Grok, DeepSeek, Kimi, a subscription you already pay for, a model on your own laptop, or the Claude CLI it finds on its own.