Recorded demo · real scan, real answers

Ask, then walk

This is the author's living room, scanned once with an iPhone. Pick a question: you get Recall's answer, then the steps the phone would speak, drawn as a route to the thing.

Ask Recall

Loading the scan…

Green: the walk, from the ring where the scan began (faint: still to walk). Dark blue disc with a pointer: you. Large dark blue marker: the thing.

You asked “Where is my watch?”

Recall says

Your grey watch is on the cabinet in the living room on shelf 1 at waist height, seen on Sunday at 19:11.

The walk

  1. Stand where the scan began, facing into the room.
  2. Turn to 11 o'clock, then about 1.1 metres.
  3. Turn to 1 o'clock, then about 60 centimetres.
  4. Turn to 11 o'clock, then about 2 metres.
  5. Arrive: Your watch should be in front of you, on the cabinet, at waist height.

Recorded output, not live: Muse Spark Open's real replies, recorded on 2026-11-03, replayed through Recall's own code. The routes come from the same planner the phone uses, on this scan's floor.

On the iPhone: AR, by voice

This page shows a recording and a scan. On the iPhone the same memory works live, in the camera and hands-free.

  1. It recognises the room

    “Scanning Living room.” Turn slowly at chest height. ARKit and a photo match on your Mac work out where you are.

  2. Labels float over the room

    Everything Recall remembers appears in the camera: found by SAM 3.1, placed by MapAnything's geometry.

  3. It asks, then listens

    “What are you looking for?” The phone taps, the microphone opens by itself, and Muse Voice Transcribe turns your words into text.

  4. It walks you there

    Muse Spark Open answers from memory. Arrows lie on the floor, and clock-face steps are spoken as you move.

  5. It checks, then asks what's next

    SAM 3.1 and LiDAR depth confirm the thing is there. “Do you want to go back, or are you done?” Muse Spark Open reads your reply.

When it listens

Only on your turn, never in the background. The phone taps, you speak, and the microphone closes when you stop talking. It isn't a live stream: your turn becomes text in about 2.3 seconds, and Recall keeps only the words.

Run this yourself

Clone the repo and run one command. No keys, no GPU, no account: the same recorded answers, served by Recall's API on your computer, and this page at localhost:3000/demo.

git clone https://github.com/coder-RT/recall && cd recall && make setup && make demo

Explore the whole scan