Glossary
Last Updated by Alex Rochira on October 29th, 2025.
Overview
This page outlines and explains Edgify's terminology
Core Components
- Agent - The main software component, installed on each edge device. Captures data through customer input and interacts with the scale/POS application.
- Trainer - Software component which hosts the AI training. Can run directly on edge devices or Edgibox.
- Edgify Dashboard - Web-hosted device management and reporting dashboard.
- Model - The AI "brain" trained using data from stores. Sits on the edge device and managed by the agent. Makes predictions and identifies loss events. Models are continuously trained using new data to improve and maintain performance.
AI Terminology
- Class - a single SKU/unique item for sale.
- Inference - Engaging the model to identify items or detect loss events.
- Sample - A single image.
- Session - A single video of a customer's shopping journey at the point of sale. Contains footage of items being scanned/not scanned, used for training our AI.
Edgify Modes
UI configurations. Controlled either through the Edgify agent, or a toggle within the application software (defined during the integration process).
- Silent Mode - The agent remains fully functioning and API communication is unchanged, but there should be no activity visible to the customer in the UI. This could mean hiding predictions, or hiding on-screen interventions. Allows Edgify to unobtrusively gather data early in deployments.
- Active Mode- The state in which predictions and interventions are visible to the customer in the UI.
- Disabled - Your software stops any communication with Edgify (Kill switch).
Device States
- Ready - The device is connected to the Edgify Dashboard and successfully sending keepalives
- Disconnected - No connection to dashboard for <24hrs
- Down - No connection to dashboard for >24hrs
KPIs
- Precision - Percentage of total transactions that were predicted correctly by Edgify's AI model
- Prediction Rate - The prediction rate is calculated as the number of predictions made divided by the total number of transactions made.
- Unknown % - Percentage of total transactions where no prediction was made due to a lower than expected confidence threshold. Unknown values are typically observed when a new item is introduced into Edgifys models or when an item cannot be identified via its respective image (i.e. camera was covered / opaque bag were used etc.)
- Mistake % - Percentage of total transactions where a customer manually choose a different label to the one that Edgify predicted. This is otherwise known as the total percentage of ‘Mislabels’ that occurred across all customer transactions.
- Recall - This is an industry standard metric for measuring the performance of an AI Model. It is calculated as: Precision X Prediction Rate
- Accuracy - This is an industry standard metric for measuring the performance of an AI Model. In Edgify's classification, the Accuracy is equal to Recall
- Top 1 - When measuring Precision, Top 1 means that Edgify was correct in the top 1 item within the predicted items list
- Top 5 - When measuring Precision, Top 5 means that Edgify was correct in at least one of the items within the top 5 in the predicted items list
Dashboard main parameters
- Status - Top 1-3, Mistake or Unknown
- Predicted at - Time of prediction
- UUID - The UUID represents the unique identity of each image. By clicking on it, you can view the saved image.
- Label - The customers product selection
- Top1 Value - Edgify's AI Recommendation
- Top1 Acc - Confidence percentage of prediction (100% = 1)
- Top2-5 Value - Edgify's 2nd, 3rd 4th, 5th recommendation
- Top2-5 Acc - Confidence percentage of Edgify's 2nd, 3rd, 4th & 5th predictions
- Certain - If Edgify has high confidence (more than a configured percentage), we mark the prediction as certain.
- Source - Refers to the source from which the item selection was made
- Device Name - This refers to the device that processed the transaction
- Cluster Name - Name of the branch to which the device belongs
- Account Name - Name of the account to which the device belongs
- Is label In Model - This indicates whether or not the item is included in the model
Dashboard User Roles
| User Role | Description | Reports | Device Management | AI Training Management | Models Management | Account Settings |
|---|---|---|---|---|---|---|
| Viewer | Data Review Settings- Account Settings (non editable) User Profile | Device Reports | Devices & Clusters | No Access | No Access | No Access |
| Annotator | Data Review Samples Sessions Settings Account Settings (non editable) | |||||
| User Profile | Device Reports | Devices & Clusters | No Access | No Access | No Access | |
| Tech | Devices Clusters Models Overview Deployments Settings Account Settings (non editable) User Profile | Device Reports | Devices & Clusters | No Access | No Access | No Access |
| Owner | Reports Devices Overview Diagnostics Clusters (Stores) Devices page only Data Review Samples Settings Account Settings (non editable) Users User Profile | Device Reports | Devices & Clusters | No Access | No Access | Has Access |
| MLOps | * Reports * Devices *Clusters *Data Review Training Models Settings Account Settings (non editable) User Profile | Device Reports | Devices & Clusters | No Access | Deploy Models | Has Access |