Version 2.3.5 | July 8, 2025

Overview

ZBrain Builder 2.3.5 introduces unified integrations, enhanced access control, refined monitoring, flexible agent memory, and adaptive knowledge graph tooling, delivering streamlined oversight, safer automation, and a smoother enterprise-grade AI experience.

ZBrain Builder 2.3.5 release overview

Together, these upgrades simplify workflow design, enhance operational efficiency, and provide enterprises with clearer visibility and control over every AI deployment.

New features

Flows & Pieces

Real-time and scheduled Zendesk triggers

Users can either select an existing Zendesk connection or create a new one during the setup process. Once configured, the associated agent should be triggered automatically whenever a new issue is created in the specified Zendesk view (a filtered list of support tickets based on certain criteria).

Key outcomes:

• Automatically triggers the associated agent as soon as a new ticket appears in the specified Zendesk view, eliminating the need for polling or manual intervention.

• Allows users to reuse an existing Zendesk connection or create a new one during setup with minimal effort.

• Accelerates triage, enrichment, and routing, reducing time-to-resolution and helping meet service-level agreements such as response and resolution time commitments.

Evaluation framework

In-place notification flow editing from Event Settings

ZBrain now supports direct editing of linked notification flows from the Event Settings screen. This feature enhances usability by allowing users to quickly update or troubleshoot notification logic without needing to navigate away from the monitoring configuration.

Navigation: Monitor → Click on any Event → Event Settings → Enable “Send Notification” → Click the Edit (✏️) icon next to the selected flow

Key outcomes

• Streamlines the notification setup process.

• Empowers users to manage automation logic with greater confidence from a single screen.

Multi-channel notification flows for Event monitoring

ZBrain now supports configuring multiple notification flows for a single monitoring event, allowing users to send alerts simultaneously across various channels. The system prevents users from adding the same flow more than once to avoid redundant alerts or conflicts. Upon meeting the evaluation trigger conditions, all attached flows are executed in parallel. This enhancement significantly improves alert reach and real-time responsiveness in critical workflows.

Navigation: Login → Monitor → Click on any Event → Event Settings → Enable “Send Notification” → Click “Add Flow” to attach multiple flows

Key outcomes

• Empowers teams to establish robust, multi-channel alert systems

• Ensures critical updates reach the right stakeholders through the platforms they use most frequently.

• Supports multiple flows for a single event.

• All configured flows are executed in parallel, ensuring timely communication across selected platforms.

Guardrails

Topic-based Guardrails with adjustable thresholds and model choice

This feature introduces advanced content moderation controls to Guardrails, enabling topic-wise filtering and model selection for enhanced enforcement of business-specific policies. When configuring or editing an App, users can now enable Guardrails that screen content by topic, Sexual Explicit, Harassment, Hate, and Dangerous Content, and set a per-filter strictness level (Block Most, with lighter options coming soon). A single evaluation model (OpenAI, Grok, or Gemini) is selected once for all Guardrail checks and shown in the UI for full transparency. Each filter (Input / Jailbreak) can have its own independent threshold selection, but the threshold applies globally to all topics within that filter.

Key outcomes

• Granular topic toggles keep apps free of hate, harassment, and explicit or dangerous content.

• Strictness controls enable teams to adjust risk levels up or down without requiring code.

• Flexible model selection (OpenAI, Grok, Gemini) optimizes cost, accuracy, and compliance.

• Safe-by-default presets safeguard new apps from the start.

RAG

Knowledge-Graph instruction tooling in Text Data Refinement

Within Knowledge → Create → Text Data Refinement, users can now:

1. Enter custom instructions to define exactly how a Knowledge Graph should be built.

2. See a tooltip warning that inaccurate instructions can skew results.

3. Click “Generate” to draft instructions, then hit Create to turn them into a prompt. The user can manually enter their own instructions to define how the Knowledge Graph should be generated.

4. Edit on the fly with a Use button that updates the prompt in the same window.

Key outcomes

• Precise control over graph construction

• Faster setup for non-experts via auto-generated drafts

• Reduced misconfigurations due to in-context guidance

• Seamless prompt tweaks without leaving the refinement screen

Agents

Selectable memory scopes for Agents

When creating or editing a Crew, you can now choose one of three memory modes for each agent:

Key outcomes

• Robust privacy and compliance control

• Seamless continuity where you need it, statelessness where you don’t

• Easier experimentation with and without long-term context

Inline edge-delete icon in the Crew graph

Users can hover over any connection between agents to reveal a trash icon; clicking once removes the edge. (The keyboard Delete shortcut still works for power users.)

Key outcomes

• Faster, more intuitive graph editing

• Visual confirmation before deleting relationships

• No more searching for hidden shortcuts

Agent-level description field

Each agent card now includes an editable description box for documenting its purpose, data sources, and instructions.

Key outcomes

• Instant clarity for teammates onboarding to a Crew

• Built-in documentation that persists with the agent

• Easier maintenance and audit readiness

Thought tracing in Crew Activity

The Crew Activity panel now surfaces each agent’s internal flow:

• Thinking – reasoning or planning text

• Action – chosen action description

• Calling – API / tool invocation details (if any)

Key outcomes

• Real-time crew activity log on the steps overview

• Full transparency into agent decision-making

• Rapid debugging of missteps or hallucinations

• Greater trust and explainability for stakeholders

Improvements

Integrations

Standardized and role-aware messaging for operators across ZBrain components

The default messages for newly invited Operators on each module (Agents, Apps, Prompts, and Knowledge Base) have been updated to provide a clearer and standardized message, with no ‘Create’ message or buttons present on the UI. This indicates that there are currently no records to view, without suggesting actions they are not permitted to take.

Key outcomes

• Improved UX consistency.

• Delivers role-aware messaging that maintains clarity and prevents confusion or misinformation among users.

CAPTCHA integration for login and user invites

To enhance platform security, CAPTCHA v3 has been integrated into the Login page and the invite user flow. This invisible verification helps protect the system against bot attacks and unauthorized automated actions while maintaining a seamless user experience.

Key outcomes:

• Improved security against automated logins and spam invites

• Invisible verification ensures smooth, interruption-free workflows

• Robust error handling with clear messages on CAPTCHA failure

• Protection without friction, triggered only during login and user invitations

Evaluation framework

Transparent scoring and evaluation insights in Monitor logs

The Monitor Logs have been significantly upgraded to deliver greater visibility and clarity into evaluation outcomes. Metric scores now display in both real-value and percentage formats (e.g., Score: 1.1 (21%)), eliminating the need for manual calculations.

Additionally, the evaluation engine dynamically selects gpt 4o and logs complete metadata, including model name, token usage, and system instructions, for every event.

To further enhance clarity, non-editable diagnostic notes now appear alongside applicable metrics requiring interpretation (e.g., Health Check), helping teams understand the scoring logic at a glance.

Key outcomes:

• A dual-format metric display combines raw scores and percentages for faster insight.

• Streamlined issue detection by identifying underperforming metrics in real time.

• Context-aware model selection ensures accurate evaluations with GPT-4o when appropriate.

• Comprehensive metadata logging improves traceability for QA, cost analysis, and audits.

• Inline metric notes provide instant diagnostic clarity without requiring log searches.

• Consistent score formatting across the platform simplifies reviews and reporting.

Apps

Centralized Bot configuration panel

The new Configure Your Bot panel consolidates all chatbot customization settings into one place, giving users full control over orchestration methods, knowledge base integration, system instructions, performance testing, and model parameters. Users can switch between Knowledge Base and Flow-based orchestration. They can attach and manage multiple knowledge sources (but with Flow, they can attach only one Flow), fine-tune AI model behavior, and evaluate bot performance using test prompts, all through an intuitive interface.

Key outcomes:

• Faster bot setup and customization with all configurations in a single location.

• Improved response accuracy through inference settings of the model and instruction settings.

• Greater control over AI behavior with manual or automated configuration options.

• Higher testing confidence with built-in Q&A-based performance evaluation.

• Enhanced bot adaptability for different use cases with flexible orchestration modes.

Redesigned Query History UI

The Query History interface has been updated with a new design and improved interaction flow. Users can now view, navigate, and manage their past chat sessions and associated contexts more intuitively, enhancing the overall chat experience.

Key outcomes:

• Simplified navigation of previous queries and chat threads.

• Improved context visibility for better session understanding.

• Faster access to past interactions boosts user productivity.

• Streamlined user experience aligned with modern UI standards.

Redesigned User Management screen

The User Management screen has been redesigned to reflect a clearer hierarchy, consistent toggle behavior, and accurate invite states. Improvements include visual alignment, avatar spacing, input and invite button styling, and enhanced interactions for menu options, delivering a more polished and intuitive experience for managing user roles and access.

Key outcomes:

• Cleaner role assignment flow with accurate hierarchy and toggle clarity

• Consistent visual and interaction design across invite actions and shared lists

• Better usability with refined hover states and overflow behavior

• Enhanced visual alignment for improved screen readability and interaction