Draw it. Capture it. Bring it to life.

Turn a hand-drawn circuit into an interactive model you can inspect, edit, and understand on Windows, macOS, Linux, or Android.

Choose your versionView the source
Holotrace
Analysis ready
Recognized circuit with resistor, LED, battery, and connecting wires
R1 resistor 96%
D1 led 93%
V1 source 98%

3 parts found

Confirm the read before continuing.

Resistor · R1
LED · D1
DC source · V1

From notebook linework to a circuit you can use.

Keep the speed of sketching. Add the structure of a digital workspace.

Start with the sketch

Take a photo on Android or bring in an existing image from your desktop.

Check the read

A local vision pass finds likely parts. Review those regions before deeper recognition.

Work with the circuit

Move components, draw connections, and switch between canvas, schematic, and parts views.

Review comes first. You see what Holotrace found before the image continues to deeper recognition.

The classifier, drawn out.

ResNet Tiny is written from scratch and trained on cropped circuit symbols. It is small enough to ship and deep enough to separate 44 classes of hand-drawn part.

ResNet Tiny architectureA 64 by 64 grayscale crop enters a strided 3 by 3 stem convolution, passes through four stages of two residual blocks each with channel widths 32, 64, 128 and 256, then a global average pool, dropout and a linear layer produce 44 class logits. Each residual block runs two 3 by 3 convolutions with a shortcut that projects through a 1 by 1 convolution when the shape changes.Ink crop64 × 64 × 1Stemconv 3×3 · s2BN · ReLU32 × 32 × 32blockblockStage 132 × 32 × 32stride 1blockblockStage 216 × 16 × 64stride 2blockblockStage 38 × 8 × 128stride 2blockblockStage 44 × 4 × 256stride 2Headglobal average pooldropout 0.2linear 256 → 4444 class logitsInside one residual blockTwo 3×3 convolutions and a shortcut. The shortcut projects through a 1×1 convolution when stride or channel count changes.inconv 3×3BNReLUconv 3×3BNReLUoutshortcut · identity, or 1×1 conv + BN on shape change
2.8Mparameters at width 32
64 × 64grayscale ink crop in
44symbol classes out
~11 MBexported weights

On the current validation split the classifier reaches 97.6% overall accuracy and a 0.82 macro F1. Accuracy is carried by the common classes; rare symbols still have far fewer examples than they need.

Fast path

ResNet Tiny labels each region the on-device pass proposed, and rejects false proposals through a dedicated background class.

Recovery path

Faster R-CNN with a MobileNetV3 FPN backbone searches the complete page for symbols the local pass overlooked. 19.2M parameters, about 77 MB.

Integration

The normalizer combines boxes, confidence, and model versions with wire geometry to build editable Circuit IR.

Every prediction keeps its source location, confidence, alternatives, and model version, so an uncertain read can be explained and corrected before simulation.

A circuit workspace that gets out of your way.

Capture at the bench. Refine on a larger screen. The same circuit stays readable across touch and desktop layouts.

Edit naturally

Pan, zoom, select, place, and wire on a workspace designed for mouse and touch.

Review before upload

See the regions Holotrace found and decide what continues to the recognition service.

See the circuit respond

Inspect connections and simple circuit behavior without redrawing the whole diagram.

One workspace. Four ways in.

Try the latest development build. New packages are published after app changes land on main.

Windows

A focused editor for reviewing captures, arranging components, and working across the full canvas.

  • Mouse and touch input
  • Full circuit workspace
  • Windows installer
Download for Windows

macOS

A universal desktop build for both Apple Silicon and Intel Macs.

  • Apple Silicon and Intel
  • Full circuit workspace
  • Universal DMG
Download for macOS

Linux

A portable x86_64 package for running the circuit workspace across distributions.

  • Portable package
  • Full circuit workspace
  • x86_64 AppImage
Download for Linux

Android

Camera-first capture and a touch-ready workspace built for the device already in your pocket.

  • Direct camera capture
  • Touch-first controls
  • Android package
Download Android APK
View build history on GitHub