An AI that can load tracks and move faders is either a great co-pilot or a terrible one. The difference is almost entirely about who's in control, and how clearly the app shows it.
Three modes
- Suggest: the agent proposes, you do everything
- Ask before acting: the agent stages an action as a card, you approve or undo it
- Autopilot: the agent mixes, and stops the moment you touch a deck
One registry of actions
We don't give the agent a separate API. Every control in the app is registered once as an action with a name, inputs and an undo. The UI calls it, the agent calls it, and the MCP server exposes it, so you can drive Audya from Claude with the same actions.
Touch a deck and you have control again.
We're starting with prep and curation (building crates, setting cues, planning an energy arc), where a mistake costs nothing. Live mixing comes later, and only behind the modes above.
