Version 2.3.6 | July 17, 2025

Overview

ZBrain Builder 2.3.6 introduces Grok 3 model support, contextual in-app help, smarter monitoring, granular jailbreak guardrails, credit-usage transparency, and a metrics-driven Apps table, enabling seamless integration, tighter governance, and data-driven oversight for enterprise AI.

ZBrain Builder 2.3.6 release overview

Together, this blends unified integrations, precision safeguards, and richer metrics into every layer of the platform. The result: faster build cycles and tighter, more transparent control over enterprise AI operations.

New features

Evaluation framework

Continuous monitoring for human-in-the-loop agents

When Human Check Required is enabled, Monitoring now records and evaluates the entire execution of the agent: initial trigger, approval phase, and final output, just as it would for any fully automated run. The log starts with 'Processing', holds while waiting for approval, then resumes, runs every configured metric (such as similarity and health check), and finishes as either 'Success' or 'Failed'. Evaluation with all assigned metrics (e.g., similarity, health check) after human approval is executed.

Navigation: Agents → Agent Settings → Additional Settings → Human Check Required

Key outcomes

• End-to-end visibility with complete logs and metrics, including during manual approval stages.

• Metric integrity—every configured check runs post approval, ensuring transparent comparisons.

Guardrails

Category-based jailbreak guardrails

Guardrails now introduce three configurable Jailbreak categories, System Override, Code Injection, and PII Access, for more granular control and security oversight. The user can enable or disable each category for Jailbreak filtering.

Model re-selection after app creation

• After creating an App, users can return to the Guardrail section and change the selected evaluation model (e.g., from Clade to Gemini or Groq). The model dropdown reflects the current selection and allows editing. Once updated and saved, the newly selected model is used for all Guardrail evaluations moving forward.

Key outcomes

• Targeted protection: Blocks override attempts, unsafe code, or PII scraping without over-filtering other content.

• Model-agnostic: Switch models post-creation; guardrails apply instantly.

RAG

Credit-usage visibility for Knowledge Graph imports

ZBrain now displays the total credits consumed by each Knowledge Graph import. Clicking ‘Details’ reveals a full cost breakdown, including knowledge base creation costs, ensuring complete transparency.

Navigation - New Knowledge Base → Text Data Refinement → Credit Usage

Key outcomes

• Instant cost insight before leaving the workflow.

• Detailed cost breakdown with drill-down access to knowledge base creation and other charges.

• Better budgeting by tracking spend in real-time and planning future imports with confidence.

Agents

Introduced instruction configuration for ZBrain Flow agents, allowing users to define and customize input parameters, expected behaviors, and output handling for Flow- based agents. This enhancement enables users to create clear, structured, and context-aware instructions to guide how Flow-based agents operate within autonomous workflows.

Key outcomes

• Customizable parameter passing clarifies how Flow-based agents should utilize input parameters and process responses, thereby improving decision-making accuracy.

• Well-defined inputs and expected actions minimize ambiguity, reducing errors during autonomous executions.

• Defined guidance on response handling facilitates smoother integration between Flow-based agents and supervisor agents.

• Optional schema definitions enable users to specify detailed input formats, thereby enhancing accuracy and providing model guidance.

Integrations

Integration with the Grok 3 model series

ZBrain now connects to Grok 3 Beta, Grok 3 Mini Beta, Grok 3, and Grok 3 Mini, as well as Twitter-tuned endpoints, via OpenRouter. All four models are available in the standard model picker and can be invoked in agents, flows, and monitoring jobs.

Key outcomes

• Broader model choice.

• Seamless Twitter enrichment.

• No-code adoption.

Context-aware documentation panel

A “?” Help icon now appears in every ZBrain module- Knowledge Base, Apps, Agents, Prompts, Monitor, and Settings. Clicking it opens a side panel with the relevant documentation for your current module (e.g., Apps, Prompts), keeping you on the same page.

Key outcomes

• Access the right documentation instantly, no searching, no tab switching.

• Consistent UX with the same Help entry point and layout across all modules.

• Higher productivity as the users stay focused in the workspace while referencing guidance.

Improvements

Apps

Expanded metadata in the Apps list

The Apps screen now shows six new columns: ID, Sessions, Avg. Response Time, Satisfaction Score, Last Updated, and Status, rendered in a table with human-readable timestamps.

Key outcomes

• Immediate insight into each app’s usage, latency, health, and recency.

• Quick filtering and triage, even on smaller viewports, without performance lag.

Evaluation framework

Intelligent event log filtering

This enhancement refines production monitoring by completely suppressing logs generated from manual “TEST” runs, ensuring that Event Logs and Dashboards capture only authentic CRON- or system-triggered activity. It also delivers polished notification templates, covering email, Slack, and plain-text notifications, that present evaluation results with a consistent layout, accurate placeholders, and a professional tone.

Finally, the Events Dashboard gains a new Refresh button that synchronizes with the Event Logs view, fetching the latest events in real-time without a page reload, allowing operators to monitor live system behavior more efficiently.

Key outcomes:

• Cleaner monitoring that filters out dev/test noise from production views.

• Sharper insights with logs that capture only real production events, boosting report accuracy.

• Lower storage prevents test-data overload in monitoring databases.