Application Note: Video - ANV089
Using Eeva with Brivo Eagle Eye Video
2026-10-16 Revision 3.0
This Application Note is designed for installers, technicians, and end-customers responsible for deploying and configuring Custom AI Analytics with Eeva and Brivo Eagle Eye Video. It outlines recommended best practices for analytics setup, camera configuration, and overall system optimization.
Eeva is an advanced video intelligence system that lets users create AI-driven events with simple natural-language prompts. It is a multi-layer model that processes the full context of on-site activity, allowing it to detect complex conditions that traditional analytics cannot.
Eeva helps organizations solve operational, safety, and security problems by identifying objects, actions, or situations unique to their environment. Instead of relying solely on prebuilt analytics, users specify exactly what they want the AI to look for, test those prompts against historical video to confirm accuracy, and then deploy them with confidence.
Users engage with Eeva via the camera interface by selecting a camera, writing a prompt, choosing an AI trigger (e.g., person/vehicle detection, object intrusion, line crossing, or periodicity), and setting the frequency. When deployed, Eeva generates custom events that are viewable in the History Browser, searchable in Video Search, or usable for alerts and automations.
The underlying workflow is simple: a trigger event occurs, Eeva reads the written prompt, cloud AI analyzes the video segment, and creates an Eeva Event if the condition is met.
Example events that can be triggered using Eeva:
Notify when a shopping cart blocks a hallway for more than 5 minutes.
Notify if a refrigeration room door is left open for more than 30 seconds.
Identify if a worker enters a warehouse aisle without a helmet or high-visibility vest.
Trigger an event when boxes, carts, or equipment obstruct a fire exit door.
Identify when someone parks a motorcycle or bicycle in car-only spots.
The prerequisites for Eeva agents are below.
Optimal camera installation is essential for achieving the best results with Eeva agents. This section outlines the recommended practices.
For Eeva to provide accurate results, position the camera to provide a clear view of the object/vehicle/person the Eeva prompt refers to. The desired minimum size is listed below.
Person Height: 1/3 of the frame Height
Object/Vehicle Width and Height: 1/4th of frame Height
At least 80% of the person/object/vehicle must be unoccluded.
For a smaller person/vehicle/object, Eeva would still work, but the accuracy would degrade.
The general rule is that users can expect good results with Eeva prompts and camera settings if a human can easily identify items by quickly glancing at the paused preview video.
For optimal Eeva performance, the system issues a warning if a camera's preview quality falls below the recommended level. Address these alerts immediately to ensure accurate analytics. If a warning appears (see example below), perform the recommended modifications:
Example Warning & Required Actions:
For Eeva to function reliably, the detected objects must meet the Minimum Pixel Requirements. If the camera is zoomed out too far, the system may issue the following warning in the Precision Person Detection settings:
Note: Change the settings to long-range mode, as some people may be too small to detect.
Required Action: Increase the optical or digital zoom to ensure the object of interest is large enough to enable accurate Eeva analysis, or click the link in the warning to switch to long range mode.
Review Eeva's camera settings in this section.
Eeva works only in the Enhanced Web Interface of Brivo Eagle Eye Video. Please switch from the Classic interface to the Enhanced interface before starting any testing. From the Classic Interface, click the button in the top-right corner to navigate to the Enhanced Interface.
To enable Eeva on your account, turn it on from the Analytics section of the specific camera you choose.
You can find Eeva Agent settings in Dashboard > Cameras > ⋮ > Settings > Analytics > Eeva. Eeva is turned on by creating an Eeva agent: click Create Eeva Agent +.
Below is an example of an Eeva prompt that creates an alert when a person enters the warehouse aisle without a helmet or high-visibility jacket. Users should describe in natural language exactly what they want to detect. The better the description, the more accurate Eeva is likely to be. Users can also choose a ready-made prompt from Select Eeva Recipe, which fills in the prompt, the label, and the trigger.
Selecting the right Eeva prompt is critical to configure Eeva for accurate event detection. A well-written prompt helps Eeva understand the scenario, objects involved, and the specific condition to detect. When formulating the prompt, keep the following guidelines in mind:
Clearly Describe the Object or Scenario of Interest
Eeva performs best for analytics involving people and vehicles, but also works well for many other object categories. Accuracy varies depending on:
Object size and visibility
Camera angle and zoom level
The complexity or granularity of the question
As a general guideline, Eeva is more reliable when the prompt revolves around visible, distinctive characteristics rather than subtle or fine-grained details.
Examples (Higher Reliability):
“Notify when a car blocks an entrance gate.”
“Alert if a vehicle is parked in the designated fire lane.”
Examples (Lower Reliability, but possible with good camera placement and zoom):
“Identify whether a pizza slice has sausage on it.”
“Detect if someone is holding a coin.”
“Determine if a person’s ID badge shows the correct company logo.”
2. Use Clear, Descriptive Natural Language
Eeva is designed to understand prompts written in human-like language rather than programming syntax. Users should write complete, descriptive sentences to describe the scenario, which often yields better results than short, fragmented prompts.
Use well-formed grammar, clear subjects, and explicit details with correct spelling.
Provide context that helps Eeva interpret the scene correctly.
Write longer, descriptive prompts (up to 2048 characters), especially when the scenario involves more than one condition.
Examples (Short, fragmented prompt):
“helmet warehouse”
Examples (Complete, descriptive prompt):
“Identify if a worker enters a warehouse aisle without a helmet or high-visibility vest.”
For ready-made prompts, see Eeva Recipes.
3. Understand Expected Accuracy Levels
With well-crafted prompts and correct camera placement, Eeva typically achieves 90% of human-level accuracy. Eeva’s human-level accuracy reflects a person’s ability to make a quick 2-second judgment on a preview frame. It is strong for broad events and visible behaviors, but limited for fine details or poor-quality imagery.
Eeva Recipes are ready-made prompts for common use cases. Selecting a recipe fills in the prompt, the label, and the trigger, which can then be edited before saving.
To use an Eeva Recipe:
In the Eeva agent, click Select Eeva Recipe next to What Should Eeva Look for?
Use Filter to search the list, or browse the recipe groups.
Select a recipe. Eeva fills in the prompt, the label, and the trigger.
Review and edit the prompt, label, and trigger as needed, then test the prompt using Test Detection.
Eeva currently analyzes events generated by Person Detection, Vehicle Detection, Object Intrusion, Object Line Cross events, or LPR events. Configure these analytics for accurate Eeva results.
For example, consider the Eeva prompt: “Identify if a worker enters a warehouse aisle without a helmet or high-visibility vest.”
In this case, the relevant trigger is a Person Detection event. For best results, Precision Person Detection should be properly configured and enabled in the Analytics Settings before Eeva is turned on, as shown below:
This configuration includes:
Defining the correct Region of Interest (ROI) so that people are detected only in the relevant areas of the camera view.
Ensuring the ROI aligns with the intended use case.
In the example above, because the goal is to detect people close to the camera, the Precision Person Detection ROI (Region 1) covers only the near-field area where detection is desired. People are detected only inside this ROI.
Eeva ROI
With the Eeva ROI, the Eeva prompt is applied only to the region selected in the ROI. To set the Eeva ROI, open the Eeva agent, click Add region in Eeva regions, and draw the region where the prompt should apply.
Note that the Eeva ROI applies only to custom Eeva prompts.
Example Scenario
Prompt: “Identify if a worker enters a warehouse without a helmet or high-visibility vest.”
An Eeva event is triggered only when the worker is within the Eeva ROI and not wearing a helmet or high-visibility vest.
Eeva supports Periodic Mode, allowing users to run analytics at fixed intervals, regardless of person or vehicle detection. This is useful for routine verification. When setting up a scheduled event, Eeva provides two key configuration options:
How frequently should Eeva run: Choose the interval at which the system should perform its checks—such as once per day, every hour, or at a custom frequency. A custom frequency is set to several minutes or hours.
Set Hours: Select the operating window for the scheduled checks. Options include Always, Work hours, Non-work hours, or Custom. Custom lets users add one or more periods, each with a day, a start time, and an end time.
These controls give users fine-grained flexibility to automate routine monitoring tasks and ensure critical operational workflows are consistently followed.
After setting up the custom Eeva event, test it to ensure the prompt works as expected.
Testing your Eeva prompt is an essential part of building a reliable and accurate agent. Iterative testing allows you to verify performance, make adjustments, and ultimately arrive at the most effective prompt for your use case. To begin, click the Test Detection button.
The Test Detection interface allows you to evaluate your prompt using:
Live Views
Historical Events, selectable from:
All Events
People
Vehicles
LPR
Intrusion
Line Crossing
Processed Images
Examples
The image below shows an example of testing on a live camera feed.
Eeva lets you freeze a live camera frame and test your prompts in the real environment, so you can validate the exact conditions under which Eeva will operate and make any necessary adjustments for optimal accuracy.
The image below shows an example of historical event testing.
Select the desired time range at the top. This example shows the Last 24 hours.
Allow the event images to load.
Use the left and right navigation arrows to browse through the images quickly.
Click any event image you want to test.
Select Analyze to run your prompt on the chosen image.
Test multiple samples to evaluate how consistently the prompt performs.
Refine the prompt using the text bar in the upper-left corner of the Test Detection UI, making small, incremental adjustments until the accuracy meets your requirements.
The image below shows the Examples tab used for testing Eeva on example images.
To test your Eeva prompt using example images:
Open the Examples tab in the Test Detection interface.
Click the upload area (Click to upload a JPEG or PNG image) and select the JPEG or PNG image to evaluate. The uploaded image appears as a thumbnail next to the upload area.
Click the thumbnail to open the image in the preview area.
Select Analyze Image to run your prompt on the uploaded image.
Review the result. Eeva marks the image as Detected or Not detected, highlights the objects it found, and shows a one-sentence explanation of what it saw.
Refine the prompt using the text bar in the upper-left corner of the Test Detection UI, then select Analyze Image again until the result meets your requirements.
The image below shows an uploaded example image in the preview area, ready to analyze.
The image below shows the result of analyzing an example image:
Eeva runs your prompt against the uploaded image, allowing you to verify how well it detects the defined condition before applying it to live or historical video.
The Analytics interface gives you full control over the Eeva prompt’s operating time period. Select Schedule when this analytic is active, then choose Always, Work hours, Non-work hours, or Custom. Work hours follow the working hours set up for your account, which may differ between users and organizations. In this example account, Work hours are Sun–Sat, 08:00–17:30.
Eeva includes a Configure Advanced Model settings option that lets users choose how many frames Eeva analyzes for each check to best match the prompt's complexity. By default, Eeva analyzes the image at the event time. Selecting Configure advanced model settings reveals AI Frame Analysis, where users can add frames from before or after the event. Each frame is defined by a number, a unit (Second(s), Minute(s), or Hour(s)), and a direction (Before event or After event). Click Add Frame to add more frames.
This is useful for prompts that involve actions or state monitoring over multiple frames, such as picking up an object.
Setting this label will make it easy to search for events in the Video Search page. Enter the label in the What label should be applied when detected? field. The label also becomes the name of the Eeva agent, and it is shown in the agent list under Eeva.
Note: The name of the label is the name of the agent. To rename an agent, change its label.
The Processed Images section in the Eeva Settings-Test Detection UI shows the response of each Eeva check. It logs images processed during both event-triggered and Periodic Mode checks, showing a blue dot when the prompt’s custom condition is met and an orange dot when it is not, allowing a quick performance review of routine verification tasks.
Users can add multiple Eeva agents on a single camera stream by clicking the Create Eeva Agent + button. This creates a new Eeva agent on the same camera. Users can create up to 10 Eeva agents per camera. Each agent will be billed separately.
Eeva can be searched using two keywords in Video Search.
Use “Eeva”, or
Filter for only one kind of Eeva event using the Label picker on the Eeva filter chip.
The image below shows Eeva events in Video Search after searching for “Eeva.”
The image below shows the Label picker on the Eeva filter chip, used to filter for only one kind of Eeva event.
Each Eeva event in the results includes a one-sentence AI-generated explanation of what was detected. The image below shows an example.
Go to View in the bottom left of the History Browser and select Eeva Event under Timeline Events.
This section shows the UI workflow of creating an alert with Eeva.
Create the action:
Go to Automations > Actions and click Add Action.
Enter an action name and select an action type, such as Notifications.
Select the notification type and the recipients.
Click Add Action.
Create the rule:
Go to Automations > Rules and click Create Rule.
Enter a rule name.
Under Conditions (If), select Eeva Event as the Alert type and select the cameras.
Under Actions (Then), select the action.
Click Add Rule.
Note: Review the Automations chapter in the User Guide for the Enhanced Web Interface or AN 96: Eagle Eye Automations and Sentinel Integration for details on how to use Automations.
What is the SKU for Eeva?
The SKU for Eeva is EN-ANA-036.
Can the Eeva SKU be applied multiple times to the same camera?
Yes, the Eeva SKU can be applied multiple times to a single camera depending on the specific
analytics features or capabilities required. Please note, each Eeva Agent will be charged separately.
How do I know how many SKUs per camera are needed?
Please refer to the Brivo Eeva Data Sheet in the Business Portal or Partner Portal for detailed
configuration rules and SKU requirements per camera model and use case.
Eeva delivers proactive, intelligent operations via custom, natural-language AI analytics. Optimal camera setup, precise prompts, and careful trigger selection ensure highly specific and reliable AI events. Eeva’s end-to-end workflow, from setup to event management, provides unparalleled precision, boosting efficiency and insight for complex operational and security needs.
This revision updates the note to match the current Brivo Eagle Eye Video interface. We updated screenshots, navigation paths, and terminology throughout to match the current interface where it differs. The main changes are:
Terminology: An “Eeva analytic” is now called an Eeva agent.
Navigation: Eeva now lives under Dashboard -> Cameras -> ⋮ -> Settings -> Analytics -> Eeva, instead of a separate Eeva Analytics settings page. Eeva is turned on by creating an Eeva agent with the Create Eeva Agent + button.
Enablement: Eeva can now be turned on from the Analytics section of the camera. Sales Engineer activation and the Private Beta evaluation period no longer apply.
Recommended analytics: Precision Person Detection and Precision Vehicle Detection are now recommended for Eeva, not required.
Triggers: The old motion/person/vehicle trigger list is now a single “When should Eeva check?” choice: Periodically (on a timer) or Event detected (Person, Vehicle, Intrusion, or Line Cross). Motion is no longer a trigger.
Fidelity and frequency credits removed: Fidelity levels 1/2/3 and the credit-based frequency system described in earlier versions of this note are no longer in the interface. Fidelity has been replaced by a simpler “Configure advanced model settings” option, covered under AI Frame Analysis.
Eeva Recipes: A new section describes Eeva Recipes, ready-made prompts that also fill in the label and trigger, and lists the available recipes.
Prompt guidelines: Examples of short, descriptive, and multi-condition prompts were added. The maximum prompt length is now 2048 characters.
Testing: Testing Modes now follow the order in the interface: Live Views, Historical Events, Processed Images, and Examples. Testing using Example Images now includes all steps, from uploading an image to reviewing the result.
Processed Images: Now part of Setting up Eeva Agent. The Processed Images section covers both event-triggered and Periodic Mode checks and shows a blue dot when the condition is met and an orange dot when it is not.
Scheduling and AI Frame Analysis: These sections now follow the order in the interface: scheduling first, then advanced model settings.
Search label: The label is also the name of the Eeva agent.
Multiple agents: Users can create up to 10 Eeva agents per camera. Each agent is billed separately.
Video Search: Eeva events can be filtered using the Label picker on the Eeva filter chip. Each event includes an AI-written one-sentence explanation of what was detected.
History Browser: The Eeva toggle now lives under View -> Timeline Events.
Creating Alerts with Eeva: Now includes step-by-step instructions to create the action, create the rule, verify the rule, and review triggered alerts under Automations -> Alerts.
ROI: The Eeva ROI
SKUs: Eeva uses SKU EN-ANA-036, available as Eagle Eye Eeva Standard Monthly (EN-ANA-036-1) and Eagle Eye Eeva Standard Yearly (EN-ANA-036-12). The previous SKU tiers (L1, L2, L3) no longer apply. See SKUs for Eeva Levels.