π Overview
The KYP.ai Connect App is optimized to run with minimal overhead:
CPU usage: ~2β3%
RAM usage: <400 MB
Disk cache: <1 GB/day per user
However, when Image Processing and Anonymization runs locally, cumulative load increases - especially in multi-session environments.
Feature | Typical Increment per User | Resulting Average Footprint |
---|---|---|
Base Agent (activity capture only) | β | 2β3% CPU |
+ Image Processing | +3β4% CPU | 5β7% CPU |
+ Anonymization & Masking | +4% CPU | 10β11% CPU |
Note: The memory limit is fixed at 1 GB; Java uses this memory over time, and during image recognition and anonymization, it is consumed faster, which in time may delay the data processing.
Solution: Off-load these workloads to the KYP.ai Server to optimize performance and scalability.
π Benefits of Server-Side Processing
π» Lower endpoint utilization β restores base 2β3% footprint.
π Predictable scaling β scales with server capacity, not user count.
π οΈ Centralized governance β unified management of OCR & anonymization rules.
β‘ Faster client log-on β reduced DLL loads and cache growth on user workstations.
π§Ύ Simplified compliance β all sensitive data handling is logged centrally.
ποΈ Architectural Comparison
πΉ Local Processing (Default)
OCR Location: β Performed on the endpoint device
Anonymization: β Performed on the endpoint device
Data Flow:
π Raw OCR content is extracted locally
π€ OCR results (text) are sent to the server in functions directly created in the KYP Connect App
π‘οΈ Anonymization is performed locally on the created functions by the KYP Connect App
β Recommended for: Single-session hosts
β οΈ Considerations:
CPU load on endpoint for OCR
πΉ Server Processing (Recommended)
OCR Location: β Performed on the server
Anonymization: β Performed on the server
Data Flow:
πΈ Screenshots captured on endpoint
π¦ Compressed screenshots are sent to the server
π OCR is performed on the captured screenshots on the server
π‘οΈ Anonymization done by the server on the functions and screenshots
β Recommended for: Multi-host sessions or shared desktops
β οΈ Considerations:
Requires screenshots collection, which takes server`s disk space and performance
π Prerequisites
π§ KYP.ai Server version
Edinburgh 2025.03.01.01
or newerπ§ Server capacity: β₯4 vCPU per 100 concurrent users. For more recommendations, please refer to the Installation requirement guide.
π Stable upstream bandwidth: β₯10 Mbit
π TLS certificates in place (see SSL Configuration section of server manual)
π§ Admin access to:
Admin Panel β Configuration β Other β Special Settings
KYP Connect App Profiles
π οΈ Step-by-Step Procedure
β οΈ Maintenance Notice
It is important to perform changes outside business hours or through a phased rollout to minimize the impact on the data collection.
π§ 1. Enable Image Processing on Server
Go to: Admin Panel β Configuration β Other β Special Settings
Enable:
IMAGE_TO_TEXT_ON_THE_SERVER_ENABLED = true
(Optional) If using an external OCR engine, change the parameters mentioned below to values that correspond to your OCR engine.
EXTERNAL_OCR_SERVICE_ENABLED = true EXTERNAL_OCR_SERVICE_ADDRESS = https://ocr.example.com/api EXTERNAL_OCR_SERVICE_TOKEN = [your_token]
π 2. Check the Anonymization configuration
Go to: Admin Panel β Configuration β Organization β Special Settings
Toggle:
ANONYMIZATION_RULES_ENABLED = true
Leave
CLIENT_DATA_ANONYMIZE = NONE
(or set toDATA
/AGGREGATED
per policy if needed)Go to: Admin Panel β Configuration β Organization β Anonymization & Data Masking β Properties
Change the Anonymization device to KYP Server:
π§© 3. Disable OCR in KYP Connect App Profiles
Go to: Admin Panel β Service β KYP Connect App Profiles β Operational
For each profile:
β Disable KYP Connect App Image to Text (OCR)
πΎ Save profile
Apply profiles to all relevant users/machines
Note: All Screenshot Collection has to be enabled in order for the OCR to be performed on the server.
π 4. Verify Data Flow
Navigate to: Admin Panel β Home β Quick Actions β Data Flow
Confirm:
π· Confirm Screenshot and OCR Data Collection
Navigate to the Screenshots and Functions item section and verify that screenshots are being captured correctly and that OCR-extracted content is present.
π Monitor Hardware Metrics Panel
Ensure that CPU usage on individual workstations has dropped to expected levels (β€3%) following the off-loading. This confirms the reduced local resource consumption.
π Validate Anonymization Functionality
Check the Anonymization & Data Masking section to confirm that rules are being applied and sensitive data is being masked as expected. Make sure to select screenshots masking in the design
π 5. Roll-Out to Production
Stage deployment:
10% β 30% β 60% β 100%
Monitor:
π Performance Guard Alerts (
AGENT_ENABLE_PERFGUARD
)π Server load via Prometheus / Grafana: CPU usage <70%
Log changes in Audit Log
π οΈ Troubleshooting Tips
Symptom | Likely Cause | Resolution |
---|---|---|
Endpoint still shows 7% CPU | Profile not fetched | Verify the profile in Connect App management, wait 15 minutes if correct and restart the KYP Connect App if still not working |
Screenshots not processed | Screenshots collection disabled or OCR service unavailable | Enable Screenshots collection and verify the OCR parameters in special settings and KYP Connect profiles |
Data not being anonymised |
| Re-enable setting & reprocess |
Server spikes to 90% CPU | Insufficient server capacity | Scale containers / add vCPU on the server |
π Rollback Plan
In case the changes need to be rolled back, so that the processing will be done on userβs workstations again, please follow these steps:
Re-enable features in KYP Connect App Profiles ( Check the parameters mentioned here)
In Special settings, set:
IMAGE_TO_TEXT_ON_THE_SERVER_ENABLED = false ANONYMIZATION_RULES_ENABLED = false
Change the Processing Device in Admin Panel β Configuration β Organization β Anonymization & Data Masking β Properties
Restart Connect Apps or wait for the configuration to be retrieved from the server ( This will take up to 15 minutes) β local OCR processing resumes
FAQ
Why should I move OCR processing to the server?
Off-loading reduces CPU usage on endpointsβespecially in multi-session or thin-client environmentsβby removing the need for local processing. It also improves performance, simplifies compliance, and allows centralized rule management.
Will off-loading affect real-time data collection?
No. Activity capture still happens in real time on the endpoint. Only the heavy OCR and anonymization processes are deferred to the server, which handles them efficiently without delaying function creation.
Is the data secure during server processing?
es. Screenshots and raw data are transferred over TLS-encrypted channels. Anonymization is handled on secure server infrastructure, and all processing activities are logged for audit purposes.