Skip to main content

Skill Studios

When you install certain skills, Aegis AI creates a dedicated studio view in the sidebar. Studios are full-screen monitoring dashboards that give you real-time insight into what the skill is doing.

Each studio includes:

  • Status strip — skill state, loaded model, device, backend, format, GPU info
  • Performance gauges — circular gauges showing inference timing (avg, p50, p95)
  • Live feed — real-time frames processed by the skill
  • Console — terminal output with stdin input
  • Config drawer — skill-specific settings

Detection Studio​

Appears when: YOLO Detection skill is installed.

The Detection Studio monitors real-time object detection across your cameras.

Status Strip​

An 8-column info bar showing:

ColumnDescription
StatusRunning / Stopped / Starting / Error with color dot
ModelThe active YOLO model name
DeviceCPU, CUDA, or MPS
FormatModel format (PyTorch, ONNX, TensorRT)
GPUGPU name if available
InferenceCurrent avg inference time in ms
ClassesNumber of object classes the model can detect
DetectionsTotal object detections since the skill started

Performance Gauges​

Three circular gauges:

  • Inference — time for the YOLO model to process one frame
  • Post-processing — time for NMS and bbox extraction
  • Total — end-to-end time per frame

Each gauge shows avg, p50, and p95 timings. A sparkline graph tracks inference time trends.

Live Detection Feed​

Per-camera slideshow showing annotated frames with bounding boxes:

  • Color-coded by class (person = green, car = blue, etc.)
  • Confidence percentage on each detection
  • Frame counter and timestamp
  • Respects Blind Mode — frames are blurred when privacy is active
  • 20-frame ring buffer per camera, cycling automatically

Detection Class Breakdown​

Bar chart showing detection counts by class (e.g. "person: 847, car: 234, dog: 12"), sorted by frequency.

Depth Vision Studio​

Appears when: Depth Estimation skill is installed.

The Depth Vision Studio shows monocular depth estimation results — colorized depth maps that visualize distance from the camera.

Status Grid​

A 3×2 grid showing:

  • Status, Model, Device
  • Backend (PyTorch, TensorRT, ONNX), Frames processed, Colormap preview

Live Depth Preview​

Per-camera depth map slideshow:

  • Colorized depth frames using the selected colormap (viridis, plasma, inferno, magma, turbo, jet, hot, cool)
  • Click to replay the 20-frame history
  • Hover to show replay button
  • Compact grid layout (up to 6 columns on wide screens)

Performance​

Same gauge system as Detection Studio:

  • Transform — depth model inference time
  • Total — end-to-end pipeline time
  • Detailed metrics: avg, p50, p95, estimated FPS, frame count

Common Studio Features​

Console​

Every studio includes a dark terminal panel:

  • Live stdout (white), stderr (red), stdin (cyan), system messages (purple)
  • Stdin input line — send commands directly to the running skill
  • Auto-scroll with manual override
  • Clear button resets all data (console, counters, and live feeds)

Config Drawer​

Click Config in the header to open a settings drawer from the right:

  • Skill-specific parameters (confidence threshold, processing interval, model variant)
  • Save button applies changes; restart the skill for them to take effect

Start / Stop​

Header buttons control the skill lifecycle:

  • Start — launches the skill process, loads the model
  • Stop — gracefully shuts down the skill and clears the running state