To help you get started quickly with the Edge Agent, several preset projects are available. These presets are ready-to-run examples demonstrating key functionalities like object detection, visual language model applications, and automated alerts. Loading a preset allows you to immediately explore a working pipeline and understand its configuration before building your own.

The Edge Agent includes several preset projects to demonstrate its capabilities across various applications. These pre-configured pipelines offer a practical way to understand how different nodes and AI models can be combined to create powerful edge AI solutions. The table below provides a comparison of the key features of these projects.

Project Name	Primary AI Model(s) Used	Key Task	Key Features
Factory Outfit Detection	OpenWord (e.g., owlv2)	Detect if personnel are wearing required attire/PPE.	Customizable detection items via JSON; L(left)/R(right) specific detection; voice alerts.
Smoke and Fire Detection	VLM (e.g., Llama-3-VILA-1.5-8B)	Detect the presence of smoke or fire in video footage.	Immediate alarm upon detection; uses prompt-engineered VLM for classification.
Clearance Space Detection	VLM (e.g., VILA-1.5-13B)	Monitor a restricted area and alert if items are placed there.	Utilizes In-Context Learning (ICL) with a reference image; Region of Interest (ROI).
Opened Door Detection	VLM (e.g., VILA1.5-3b), NanoDB (CLIP)	Monitor door status; alert if open too long; identify attire of passersby.	Two-step alert logic; image capture on event; multimodal database search.
Adaptive Door Status Detection	VLM (e.g., VILA-1.5-13B), NanoDB (CLIP)	Enhance Visual Question Answering (VQA) model accuracy using RAG.	Multi-step process (error tagging, DB learning, intelligent retrieval); dynamic context enhancement.
OpenWord FineTune	OpenWord (e.g., YOLOWorldV2_V8X)	Improve detection accuracy of an OpenWord model via training.	Utilizes user-annotated data (LabelMe); involves a model training cycle.

⬆️ Top

---

Loading and running the preset projects in the Edge Agent is straightforward. These presets provide a quick way to see the agent's capabilities in action.

1. Ensure prerequisites are met:
   • The Edge Agent is installed and running on your Jetson device.
   • You can access the Edge Agent UI via your web browser at https://IP_ADDRESS:8050.
   • Essential data like demo videos and images required by the presets have been downloaded and placed in the correct directories (e.g., /ssd/jetson-containers/data/videos/, /ssd/jetson-containers/data/images/, /ssd/jetson-containers/data/datasets/).

2. Loading a preset project: Each example project typically has a corresponding preset name (e.g.,DOOR_ALERT_WEBRTC_ADVAN, ATTIRE_DET_WEBRTC_ADVAN).There are multiple ways to load a preset:

   • Using the "Agent" menu (recommended for full pipelines):
     1. In the Edge Agent UI, click on the "Agent" button in the top-right corner.
     2. Select "Load" from the dropdown menu.
     3. A list of available saved presets (JSON files from the jetson-containers/data/nano_llm/presets directory) will appear. Select the desired preset name corresponding to the project you want to run.
   • Right-clicking in the node editor (often for subgraphs):
     1. Right-click anywhere in the main Node Editor canvas.
     2. A context menu should appear. Look for options like "Load," "Insert," or a submenu for presets/subgraphs. Select the desired preset or subgraph (e.g., a pre-configured ASR or TTS module).
   • Using the --load flag at startup (For advanced users):
     1. When starting the Edge Agent from the command line, you can use the --load option followed by the path to the preset JSON file to pre-load a specific pipeline.

3. Running the project:

   • Once a preset is loaded, the corresponding pipeline of nodes and their connections will automatically appear in the Node Editor.
   • The project will typically start running automatically. Data will begin to flow from the VideoSource (if configured with a video file or camera) through the connected nodes.
   • Observe the output:
     • Check the VideoOutput panel in the UI. This will display the video feed, often with visual overlays like bounding boxes for object detection, text from VLM analysis, or status messages.
     • Listen for audio alerts if the project uses a Text-to-Speech (TTS) module like PiperTTS.
     • Monitor the Terminal panel for log messages, detailed text outputs, or error messages.
     • Interact with node-specific UI elements if the project uses them.

⬆️ Top

---