Version 2.3.4 | June 26, 2025
Overview
ZBrain Builder 2.3.4 delivers faster flow building, streamlined operational oversight, safer and more expansive automation, refined monitoring, and greater agent transparency. Together, these upgrades help teams develop, operate, and monitor enterprise-grade AI solutions more responsively, efficiently, and intuitively. The release packs several new features and targeted improvements that further elevate user experience and productivity.
ZBrain 2.3.4 release overview
Flows & pieces
Auto-scrolling Flow interface
Automatically scrolls the interface during component dragging for a smoother workflow editing.
Integrations
Standalone executable of Resume Screening Agent
Downloadable .exe file for Resume Screening Agent with complete resume review functionality.
Expanded file type support, file type validation and preview of file.
Adds support and validation for new file types (PNG, JPG, JPEG).
Evaluation framework
Real-time processing status
Shows live status updates for triggered events, reducing workflow uncertainty.
Bulk log deletion
Deletes multiple event logs to improve productivity.
Health check metric
Validates app/agent uptime through HTTP checks for easy monitoring.
Guardrails
Parameter-specific configurations
Generates an individual .yml
file per parameter, simplifying configuration updates and audits.
Dynamic query processing
Context-aware validation dynamically applies relevant Guardrail rules.
Enhanced debugging & observability
Detailed logs for easier debugging of rule execution and response filtering.
RAG
Locked RAG definition
Locks the RAG definition at initial KB creation to prevent misconfigurations.
Agents
Google search integration
Adds real-time web search capability to agents.
ZBrain Flow integration
Embeds custom ZBrain Flows within agents to streamline automation.
External package support
Enables integration of external packages directly within agents.
Agent type visibility
Clearly displays agent types for easier identification and management.
Usability enhancements
Adds message sending shortcuts and safer instruction editing for Test Crew.
SaaS
Seamless scrolling for Knowledge Storage
Enables continuous scroll for smoother browsing of extensive records.
These enhancements collectively streamline workflow creation, boost operational efficiency, and deliver greater transparency for enterprise-grade AI deployments.
New features
Flows & Pieces
Auto-scrolling on the Flow interface during component dragging
The Flow interface now automatically scrolls whenever a component is dragged toward any edge, allowing it to move in all directions. Precise snap-to-grid placement is preserved, and the feature works seamlessly with existing zoom and pan capabilities, eliminating the need for manual scrolling on large or complex flows.
Key outcomes:
Auto-scroll activates when dragging within the interface edges.
Speeds up flow editing and provides a smoother design experience, especially when working on flows that extend beyond the visible screen area.
Maintains precise positioning and snapping of components.
More intuitive, efficient, and seamless experience for users building complex workflows.
Integrations
Standalone executable of Resume Screening Agent
Users can now export the Resume Screening agent as a single, self-contained .exe (no installer, no extra folders). Navigation: Resume Screening Agent › ⋮ (More) › Download Agent

Key outcomes
Delivers complete resume screening functionality without internet access, preserving full feature set.
Evaluation framework
Monitor
Real-time status for agent events
This feature adds a “Processing” status to the monitor dashboard for agent events that have been triggered (e.g., by a cron job) but have not yet posted logs. This provides real-time visibility into execution progress before results are finalized. Once logs are posted, the status should automatically update to “Success” or “Failed.”
Navigation path: Monitor → Events → Last Status(Column)

Key outcome:
The processing status confirms an event is in progress, removing guesswork when monitoring large or scheduled workflows.
Bulk delete for monitor logs
Users can now select and delete multiple event logs at once from the Monitor Events screen. This replaces the previous one-at-a-time deletion model with a faster, more scalable option.
Navigation Path: Login → Monitor → Events → (Event Checkboxes)

Key outcome:
Improves productivity and log hygiene for users managing high volumes of events.
New evaluation metric – "Health Check"
Users can now include the Health Check metric in evaluation from the Non-LLM Metrics suite. It verifies that an agent or app is reachable by checking the HTTP response status and can trigger notifications when they are configured.
Navigation Path: Monitor → Events → (Click on any ‘Event’) → Event Settings → Add Metric → Non-LLM Metrics → Health Check

Key outcome:
Offers a simple yet powerful monitoring metric to validate system uptime and API responsiveness, without requiring LLM-based evaluation or complex configuration.
Guardrails
Parameter-specific configuration files
Each monitored parameter (such as input and jailbreak) now has its dedicated .yml
file (e.g., input.yml
, jailbreak.yml
). These files are generated and versioned independently and stored in a private S3 bucket. Only users with required credentials can access those files.
Navigation: App → Create App → Add guardrail parameter → Select Guardrails → Enable as per requirement → Complete App Creation → Test Bot Performance

Key outcome:
Improves maintainability and makes it easier to audit or update specific evaluation rules without impacting the entire configuration.
Filters data based on context-specific rails through the Guardrails service.
Dynamic query processing via Guardrail service
The query engine detects the context (e.g., input vs. output) and sends the relevant .yml
reference to the Guardrails service. The Guardrails service will then process and filter responses accordingly.
Key outcome:
Enables context-aware response filtering that returns a validated response to the main flow.
RAG
RAG definition locking based on the initial KB definition
The Data Refinement Tuning page now shows only the RAG mode defined when the Knowledge Base was created, providing an intelligent UI behavior on the page. If a KB is initially created using Knowledge Graph, the Vector Store option is hidden for all subsequent document uploads under that KB and vice versa.

Key outcome:
Enforces consistent RAG architecture within a KB, avoiding misalignment or mixed retrieval strategies.
Reduces user confusion and prevents configuration errors during document ingestion.
Agents
Inclusion of Google Search in Agent default tools
This feature adds real-time search capability to agents, allowing them to retrieve current information directly from Google.
Navigation path: Define Crew Structure → Select Agent → Agent Tool → Default Tools-> Google Search

Key outcome:
Agents gain real-time web access, enhancing their ability to deliver timely and relevant responses.
ZBrain Flow integration with Crew Agent
ZBrain Builder now integrates ZBrain Flows into Crew Agent, enabling users to add flow-based agents. Agents can trigger a ZBrain Flow and use its response to guide their next actions.
Navigation path: Define Crew Structure → Add Agent → Add from Library → ZBrain Flow


Key outcome:
Seamlessly integrates advanced internal workflows into agent logic, enhancing automation and flow-based decisioning.
Addition of packages in the Crew Agent Tools
Users can now add external dependencies (e.g., npm packages) directly to an agent’s tool. On clicking ‘Add’, the latest version of the package will be automatically added to the tool's list of dependencies. This supports automatic version resolution and easy rollback.
Navigation path: Define Crew Structure → Select/Add Agent → Agent Tool → New Tool / Select from My Tools → Add Package

Key outcome:
Improves tool extensibility and supports custom agent logic through external libraries.
Visibility of Agent Type in the interface
The agent type (Supervisor, Autonomous, or ZBrain Flow) is now clearly displayed in the agent interface in an agent crew structure.

Key outcome:
Enhances crew transparency and helps users quickly identify the role and behavior of each agent.
Improvements
Integrations
Expanded upload file type support in the agent dashboard
ZBrain now supports image formats (PNG, JPG, JPEG), in addition to the existing formats: PDF, TXT, CSV, JSON, DOCX, PPTX, and XLSX.

Key outcome:
Supports a wider variety of input data without requiring pre-processing or format conversion.
Smart validation and feedback
Each uploaded file is validated for integrity and supported file type.
Invalid or corrupt files are blocked with clear error messages.
Unsupported formats (e.g., .exe, .zip) are automatically rejected with "Unsupported Format" warnings.

Key outcome:
Reduces user errors and ensures a smooth file submission experience.
Enhanced UI experience with file previews
Image files show a thumbnail preview or filename confirmation after upload. Other files (e.g., DOCX, PDF, CSV) display their filename for quick verification.
Key outcome:
Provides visual confirmation that the correct files were uploaded, improving confidence and efficiency.
Reliable task execution across all supported formats
Uploaded files (across supported types) are correctly processed during task submission. The pipeline handles input ingestion based on file type, ensuring seamless execution of the agent.
Key outcome:
Guarantees consistency in agent behavior regardless of input format.
SaaS
Seamless scrolling for knowledge storage
Automatically loads more storage items as the user scrolls toward the bottom of the list. Seamless scroll experience with no page reloads or buttons.
Navigation path: Settings → Billing → Knowledge Storage

Key outcome:
Provides a faster and smoother browsing experience for users managing extensive knowledge storage records, thereby enhancing data accessibility and overall user efficiency.
Guardrails
Enhanced observability and debugging
System logs now include configuration details, the response before and after filtering, and any errors encountered during rule parsing or execution.
Key outcome:
Provides full transparency for auditing and debugging, making it easier to trace filtering outcomes and diagnose misconfigurations.
Agents
Usability enhancements
Quick message shortcut: Users can use Ctrl+Enter to send a quick message in Test Crew.
Instruction editing: Deleting an instruction now clears only the text, not the entire agent from the Crew.
Key outcome:
Improves testing efficiency and reduces accidental agent deletions during edits.
Last updated