Skip to content

Getting Started

Detect Edge is a mobile app that turns a Nordic Semiconductor Thingy:53 into your own detector. You describe what you want to detect, record examples with the device, and Detect Edge trains a model from them. You test the model live in the app, then deploy it to the Thingy:53, where it runs on the device and your phone notifies you when it detects something.

The steps, in order:

  1. Sign in and create a project.
  2. Add and connect your Thingy:53.
  3. Record examples for each label.
  4. Test the model.
  5. Deploy it to the device and turn on notifications.

Recording uses the Thingy:53’s accelerometer, so Detect Edge suits things you can feel as motion or vibration: a door closing, a knock, an object being picked up, a machine running.

On first launch, create a Detect Edge account or sign in with email, Google, or Apple. The app asks for Bluetooth permission to find your Thingy:53. On Android it also asks for location permission, which Android requires for Bluetooth scanning; Detect Edge does not collect your location.

Everything you record belongs to a project. To create one, open the project menu at the top left of the Record or Data tab and choose to create a new project.

Describe what you want to detect in plain language, for example “tell me when the garage door closes”. The app may ask a follow-up question or two. It then suggests a title, a description, and the labels to record, each with instructions. Review them and create the project. Labels are set when the project is created.

Open the device menu at the top right and add your Thingy:53. Make sure it is switched on and nearby; it appears in the list once the app finds it. Add it to your project, then connect.

A new Thingy:53 ships with factory firmware, and Detect Edge needs its own. If the app asks to update the firmware, do that first: it takes a few minutes and happens over Bluetooth. The same device can be added to several projects.

With a project selected and a device connected, open the Record tab. The card at the top shows the label to record, how many examples you have, and what to do.

  • Choose the label. “Auto” picks the label that needs examples most. Tap a label to record that one instead.
  • Press Start, do the thing once, and press Stop. A 3-second countdown gives you time to get ready; you can turn it off with the switch under Start.
  • Review it. After you stop, the recording’s chart is shown. Choose Good to keep it or Bad to discard it.

Detect Edge checks your recordings as you go: a setup check after 3 examples of a label, a model check after 8, and a starter set at 15. If something looks wrong, the app tells you what to fix.

You don’t choose which recordings are for training and which are for testing. Every fifth recording of each label is set aside automatically for checking the model.

Once every label has at least 8 examples, Test model appears. Detect Edge trains a model on your recordings and shows its predictions live as you move the device. Use this to check that your labels can be told apart before recording more.

You don’t start training yourself. It happens when it’s needed.

When the starter set is complete (15 examples of every label), Deploy to device appears. Deploying sends the model to your Thingy:53 over Bluetooth, which takes about a minute. Keep the device close to your phone while it runs.

After that, the model runs on the Thingy:53 itself. In the model test you can switch between Cloud and On-device to compare them.

Go to Settings, then Notifications, then Detections, and turn on “Notify me about detections”. Your phone then watches devices running your models and notifies you when one detects a label, including when the app is in the background. You can choose which labels notify you.

Your phone needs to be within Bluetooth range of the device for notifications to arrive.

The Data tab shows your recordings grouped by label, with a toggle between training and testing recordings. Tap a recording to view it. You can delete single recordings, or all recordings of a label from the menu on its header.

Export. Tap the share icon at the top right of the Data tab and choose Export CSV. You get a ZIP of your labeled recordings as CSV files, which you can open in other tools.

Privacy. Your projects are private. See the Privacy Policy for details.

Contributing (optional). In Settings, under Data sharing, you can choose to contribute recordings to open datasets that help everyone’s detectors work from fewer examples. Contributed recordings are published without your name or email under CC BY 4.0. It is off unless you turn it on.

The Settings tab has your account, your storage quota, notifications, data sharing, and help. Device settings (the device’s name, firmware updates, and restoring the factory firmware) are in the device menu, next to each device.

If you have questions, a bug to report, or feedback, go to Settings and look under Help for support chat, feedback, and our Discord.