Workspace & Core Tools

Setups, Layouts, and the Activity Bar

A setup saves its plugins, Activity Bar options, run groups, and layouts. Load a packaged or custom setup from View → Load Setup or Preferences → Appearance → Setups. Use Save Current Setup to create a custom setup; packaged setups cannot be overwritten or deleted.

The Activity Bar switches between layouts inside the active setup. A layout saves the visible panels and their arrangement without changing which plugins belong to the setup.

  • Right-click empty Activity Bar space to create a layout or run group.
  • Right-click a layout to save the current arrangement, edit its name or icon, move it, or delete it.
  • Use View → Activity Bar to show or hide the bar.

Run Groups

A run group starts and stops up to six selected tools together from one Activity Bar tile.

Step Action
1 Right-click the Activity Bar and choose New Run Group.
2 Name the group and select its tools.
3 Click the group tile to start or stop the selected tools.
4 Right-click the tile to edit, rename, or delete the group.

Deleting a run group does not delete or unload its tools.

Panels

  • Float moves a docked panel into its own window.
  • Close offers to hide or unload the panel.
  • View → Panels shows or hides loaded panels.
  • View → Panels → Float All Panels / Dock All Panels changes all loaded panels at once.

Preferences

Appearance

Choose language and theme, then load, save, or delete setups. Language and theme changes apply immediately.

Hardware

  • Compute GPU: used for model discovery, inference, OCR, and TensorRT engine building. Auto lets Helios choose.
  • Capture Audio: plays a matching capture-card audio input through the default PC output.
  • Fuser Output: enables the separate Fuser display and lets you identify the selected monitor.

Python

Search for installed Python environments or choose an executable. Helios requires Python 3.10 or newer for Python-based features and shows whether the selected environment is valid.

Account

Sign in with Discord or Login with Email. Email sign-in uses a one-time link and needs no password. Keep Helios open, check your inbox and junk folder, and only approve an email whose verification code exactly matches the code shown in Helios. Sign-in finishes automatically after confirmation.

The signed-in page shows your Input Sense Account ID and age. Use the copy button to share your account ID with a resource author who needs to grant you access. This 21-digit ID is different from your Discord ID.

Use Link Email or Link Discord to add the other sign-in method to your current account. Publishing, managing resources, and integration keys require both a linked Discord account and a verified email. Browsing and receiving resource access do not require both methods.

Unlinking a sign-in method requires confirmation through your verified email and signs out other account sessions. Your Input Sense account and resource ownership remain. Keep another sign-in method linked so you can still access the account.

Other

  • Enable Developer Mode: shows publishing, GPC3 compilation, authoring controls, and the direct input-to-output passthrough switches.
  • Disable Overlay Drawing: shows raw video in Video Display while script, inference, and OCR processing and results continue normally.

Developer Mode is optional for normal Helios use. Enable it when publishing or authoring content, or when a physical input should write directly to Controller Output or KBM Output without a processor such as GPC 3.

Engine Builder

Open Tools → Engine Builder to manage downloaded weights. With a compatible NVIDIA/TensorRT Compute GPU, the dialog also imports ONNX/ENNX weights, reviews model information, builds TensorRT engines, and manages separate source and built-engine lists.

With a DirectML or other non-TensorRT selection, Engine Builder shows the Downloaded Weights list without TensorRT build controls.

Before building:

Step Action
1 Select a compatible NVIDIA TensorRT GPU in Preferences.
2 Configure a Python environment with the required TensorRT packages.
3 Confirm the model task is detect, pose, or segment, then choose the input size and FP16 or FP32 precision. Use the ModelOpt FP16 choice only for a ModelOpt mixed-precision ONNX source.

Builds can take several minutes. The status area identifies TensorRT and CUDA compatibility. An engine made for a different GPU or TensorRT environment may need to be rebuilt locally. Building protected ENNX weights also requires sign-in, access to the model, and an internet connection.

Select a built engine and click Rebuild to use its existing input size and precision. Rebuild is available only while the matching original source is present and still belongs to that engine. If the source was moved, replaced, or deleted, import the correct original weights and build again.

Dataset Capture

Open Tools → Dataset Capture to save JPEG or PNG frames from the active video source.

Available controls include output folder, optional crop, JPEG quality, minimum interval, maximum image count, maximum folder size, and these triggers:

  • while a selected controller button is held;
  • when one of the listed inference class IDs is detected;
  • at a fixed time interval.

When multiple triggers are enabled, any one can capture. Button and time triggers can save raw video without a running inference script; a class trigger requires active inference.

Updates

Helios checks for updates after startup. Use Tools → Check for Updates to check again manually. Follow the updater prompts and let it finish before reopening Helios. If a download cannot be saved or verified, check free disk space, folder/antivirus access, the internet connection, and the Windows clock before retrying.