Version 2.4.1 | Sept 22, 2025
Overview
ZBrain Builder 2.4.1 release focuses on transparency, security, and control. Users gain greater clarity in monitoring and auditing flows with improved traceability, consistent naming, and flexible run analysis options. Enhanced data handling ensures faster navigation across large datasets and more reliable knowledge base creation, supported by smarter chunking and OCR-driven content enrichment. Strengthened security safeguards sensitive credentials, while Human-in-the-Loop execution empowers users to intervene in agent workflows for higher accuracy and trust. Together, these updates deliver a platform that is more intuitive, secure, and aligned with real-world operational needs, helping teams work with confidence and efficiency.
ZBrain Builder 2.4.1 release overview
Flows & Pieces
Unique identifier column in the Runs table
Provides a unique identifier to parent and child flows, enabling users to trace, search, and differentiate specific Flow runs
Auto-sync Run names with updated Flow names in the Runs view
Ensures consistency and clarity by keeping Run names aligned with the latest Flow name across all views
Open Flow Run details in a new tab directly from the Runs list using standard browser actions
Provides flexibility to analyze multiple runs simultaneously without losing the Runs list view
Expanded date filter added in the Runs view
Gives users finer-grained control over run history filtering, enabling quick access to recent execution records
Integrations
Mask sensitive fields like API keys and access credentials when adding models in both the LLM and embedding sections
Protects confidential information from exposure, ensuring secure model setup and stronger data security
RAG
DeepDoc Parser integration for knowledge base creation with smart chunking and OCR-based data extraction
Enables more accurate and reliable knowledge retrieval by preserving contextual chunks, extracting structured tables, and generating OCR-based descriptions for non-text content
Agents
Pause agent execution at predefined steps with Human-in-the-Loop (HITL) for real-time user input
Gives users control to review, edit, or approve intermediate outputs, ensuring more accurate, trustworthy, and flexible agent workflows
New features
Flows & Pieces
Unique identifier column in the Runs table
ZBrain Builder now introduces a unique Identifier column in the Runs table to link parent and child flow executions. Users can pass their own identifier when triggering a parent flow (via Postman or HTTP), which automatically propagates to child flows. The new column and search option make it easy to trace, debug, and monitor complex executions with full visibility.
Navigation: Flows → Runs
Key outcomes
Users can easily trace related parent and child runs under a single identifier.
When a flow fails, users can identify exactly which parent or child run encountered the issue.
The new search bar speeds up locating specific runs in environments with multiple interlinked flows.
Even when Run IDs are missing, unique identifiers ensure seamless linkage across executions.
Automatic synchronization of run names with updated Flow names in the Runs view
ZBrain Builder now ensures that Flow run names in the Runs view sync with the latest Flow names, eliminating confusion when flows are renamed. Runs remain tied to permanent Flow IDs for traceability, while the UI always shows the most current name for clarity.
Navigation: Flows → Runs
Key outcomes
Users can reliably track Runs without being confused by outdated Flow names.
Simplifies auditing, debugging, and analysis by ensuring Runs always align with the latest Flow names.
Original Flow IDs remain unchanged, ensuring historical and backend accuracy.
Opening the Flow Run details in a new tab
ZBrain Builder enhances navigation flexibility in the Runs list by introducing the ability to open Flow Run details in a new browser tab using standard actions like 'Ctrl+Click' or right-click → Open in new tab. Previously, Run details could only be viewed in the same tab, which caused users to navigate back and forth when comparing or reviewing multiple runs.
With this update, users can keep the main Runs list open in one tab while reviewing detailed execution logs, token usage, and status information in another, making it easier to analyze multiple runs side by side. This update preserves the existing functionality—users can still click directly on a run to view details in the same tab if preferred. The new behavior adds flexibility without changing familiar workflows.
Navigation: Flows → Runs→ Select a flow-> Ctrl+Click/Flow Run
Key outcomes
Users can review multiple Flow Runs simultaneously without losing context.
The Runs list remains intact in the original tab, avoiding repetitive back-and-forth navigation.
Faster comparison of parent and child runs or sequential executions by viewing them in parallel tabs.
Agents
Human-in-the-Loop (HITL) for crew agent execution
ZBrain Builder introduces Human-in-the-Loop (HITL) controls at the agent level, enabling users to pause agent execution at predefined steps, review intermediate outputs, and decide whether to edit, re-run, or approve results before passing them downstream. HITL can be toggled on or off for individual agents within a crew, providing teams with the flexibility to apply it only where validation is critical. When enabled, agents automatically pause after completing their task and present conversational feedback prompts to the user. Users can approve results to continue the workflow, or modify inputs and trigger a re-run before execution moves to the next agent.
The feature is supported across all models and is implemented under the Google ADK framework. The HITL workflow integrates seamlessly into existing agent dashboards and crew structures, with controls clearly marked for pausing, resuming, and editing at each step.
Navigation: Agents → Create Agent Crew-> Define Crew Structure → Add Agent-> Enable/Disable Human in loop
Key outcomes
Users can validate and refine intermediate outputs before final results are shared, ensuring controlled response.
Early intervention reduces the risk of flawed or incomplete results propagating through the workflow.
HITL can be applied selectively to critical steps without interrupting end-to-end automation unnecessarily.
Provides visibility into agent reasoning and intermediate outputs, building user trust in automated workflows.
Supports real-time adjustments, making workflows more responsive to nuanced or evolving requirements.
RAG
DeepDoc Parser for knowledge base creation
ZBrain Builder now integrates the DeepDoc Parser into knowledge base creation, delivering advanced document processing capabilities for greater accuracy and richer retrieval. This enhancement ensures that documents are no longer split arbitrarily but are contextually chunked, while also supporting OCR-based extraction for non-textual content and structured parsing for tabular data.
This feature significantly enhances search precision and knowledge grounding within ZBrain apps and agents, reducing information loss during ingestion and enabling richer, context-aware responses.
Navigation: Knowledge → New Knowledge Base-> Data Refinement Tuning-> Chunk Settings->DocType Chunking
Key outcomes
Smarter chunking keeps related sections intact, improving retrieval accuracy and response quality.
OCR-based extraction captures and stores descriptions of images and graphs for richer context.
Structured tabular extraction preserves rows, columns, and headers in full fidelity with HTML linkage.
Combined enhancements deliver more precise, context-aware, and comprehensive answers.
Users benefit from a smoother setup process and higher-quality retrieval, especially when working with complex documents that include tables, images, or varied formatting.
Improvements
Integrations
Masking sensitive fields in model setup
ZBrain Builder now masks sensitive fields such as API keys and access credentials when adding models or embedding configurations. Previously, these inputs were displayed as plain text, risking accidental exposure. With this enhancement, all API keys and secret tokens across providers—including OpenAI, Azure OpenAI, Google Generative AI, Groq, Bedrock Claude, and Custom models—are securely masked in both the LLM and embedding sections.
Key outcomes
Prevents accidental disclosure of sensitive API keys and credentials.
Applies masking across all supported providers and model types.
Allows users to set up models confidently without exposing confidential information.
Enhanced preset date filters in Runs view
The Runs view has been improved with additional preset options, giving users greater precision when filtering execution data. Alongside the existing presets (e.g., last week, last 6 months), users can now quickly apply precise time windows, including last 15 Minutes, 30 Minutes, 1 Hour, 6 Hours, 1 Day, and 3 Days, without manually entering a custom range.
When selecting a preset, the Runs table instantly refreshes to display only the executions that match the chosen timeframe, maintaining consistency with the existing filtering workflow. This enhancement streamlines monitoring and troubleshooting by making it easier to isolate recent executions and identify issues faster.
Navigation: Flows → Runs → Pick a date range → Select Preset
Key outcomes
Provides quicker access to recent execution data without manual input.
Enhances troubleshooting efficiency by narrowing the scope to precise timeframes.
Improves monitoring workflows by aligning filtering with real-time operational needs.
Reduces friction in navigating large datasets, especially when diagnosing recent failures or performance issues.
Last updated