Conversational Dialog

CMI is a conversation, not a search box. You talk; it works out which kind of search each turn needs; and you build from what comes back.

One input for everything: type or dictate a brief, attach a script, image, or video, toggle reasoning, and send.

It picks the method, you just talk

The chat reads what you bring and routes it: Precision for a nameable sound, Narrative for a scene, Similarity for a reference track. You never pick a mode yourself; while it works, the loading bubble shows which one it chose.

And when it doesn't yet know the one thing that decides the music (the specific moment in the story the music has to act on), it asks before searching. That question comes back throughout: it is what turns a vibe and a setting into a usable cue.

Build on what comes back

Put everything you know into the first prompt rather than feeding it in piece by piece; a full brief up front often gets you there without refining at all.

When you do refine, the session carries context, so the conversation compounds. "Same vibe but slower", "more electronic", "drop the strings" refine the current result instead of starting over. CMI stays in one mode until you clearly change topic; the New Search button clears the slate. Nothing resets on its own.

Keep refinement short, though. If three or four rounds haven't landed it, a fresh New Search from a new angle usually beats pushing the same thread further.

Speak, type, or attach

• Type or speak. Dictate a turn with the mic button instead of typing.
• Attach. Add a script, a briefing, an image, or a video, and CMI folds it into the conversation: a whole document becomes a list of moments; an image or video becomes a scene.

Reasoning: faster or sharper

A toggle sets how hard CMI thinks before it routes:

• Off: noticeably faster.
• On: more accurate, and it matters most in non-English languages, where routing can otherwise misfire.

So reasoning defaults to on for non-English, off for English.

The reasoning toggle: the brain icon in the input bar.

Smart Start

Auditioning a result shouldn't mean scrubbing through a track. With Smart Start (on by default) the player opens each track at a musically relevant moment rather than at bar one, so a few seconds is usually enough to grasp its essence and judge whether it fits. Turn it off in the player to hear tracks from the top.

Smart Start, in the player bar: on by default.

If a single message carries several distinct needs, CMI doesn't pick for you; it lists them. See Multi-Search.