Mount the iPhone. Your robot grows a brain.
The phone rides on the build as a perception and reasoning module. SPIKE keeps motor timing fast and deterministic. The iPhone adds camera, motion, voice, and AI context. The line between them stays explicit.
Phone as sensor stack
Six new senses the SPIKE hub does not have.
Every sensor on the iPhone becomes a stream the assistant can subscribe to. Each one shows up in the build context with a clear name, sample rate, and what it's for.
Camera
Object detection, color tracking, line following from above
Microphone
Voice intent and ambient sound classification
Speaker
The robot can speak back: explanations, warnings, confirmations
Accelerometer
Detect tipping, hard stops, surface changes
Gyroscope
Heading and rotation, including drift correction
GPS / Network
Coarse on-device context, kept on the phone and never shared
The split
Two brains, working at the speeds they're each good at.
The hub runs tight motor loops. The phone reasons about goals. We don't pretend the LLM can do real-time control — and we don't pretend the hub can have a conversation.
iPhone brain
~1s loopsSPIKE brain
~10ms loopsWhat mounted mode won't do.
The clearer the limits, the more useful the freedoms. Here's exactly where mounted mode draws the line.
Override motor stops
The LLM cannot bypass stop conditions set in MicroPython. Period.
Stream silently
Camera and mic only run when a skill explicitly subscribes, with a visible indicator on screen.
Send AI requests without consent
Every new AI feature asks once, the first time it could matter.
Drive when the kid isn't watching
A live operator view is required to start any mounted skill.