# Version 2.4.5 | Dec 29, 2025 ## **Overview** ZBrain Builder 2.4.5 delivers major upgrades to agent orchestration, knowledge management, and model flexibility. This release introduces multi-document support in Agent Dashboards, memory-based inference for smarter follow-ups, and seamless compatibility with the Claude 4.5, GPT-5.1, and Gemini 3 models. Users can now import external agents from AWS Bedrock and Azure Foundry, review and refine KB chunks before ingestion, and manage piece versions in Flow with greater control. Together, these updates reinforce ZBrain’s adaptability, precision, and enterprise-readiness across AI applications. #### ZBrain Builder 2.4.5 release overview | Component | Capability | What it delivers | | -------------- | -------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Flows & Pieces | Copy and paste for workflow pieces | Delivers rapid duplication of logic blocks across workflows, accelerating build time. | | | Version control for workflow pieces | Enables precise management of step versions within Flows, simplifying rollbacks, upgrades, and iterative app development. | | | Claude 4.5 Sonnet and GPT-5.1 support in Ask AI Model | Empowers users to harness state-of-the-art language models for enhanced reasoning, summarization, and content generation | | | ChromaDB support for Knowledge Base | Enables creation of efficient, lightweight retrieval systems using ChromaDB, with seamless app integration for streamlined knowledge testing. | | | New model support: Gemini-3-Pro-Preview in the Ask AI Model piece | Supports Gemini-3-Pro-Preview in the Ask AI Model step, expanding model flexibility and enabling users to harness Google’s latest-generation LLM for enhanced reasoning and content generation | | | Document Intelligence and Multimodal Document Intelligence in ZBrain Flows | Enables AI-powered extraction from a wide range of public document formats using URLs (PDFs, images, text files, Office docs, etc.) | | Agents | AWS Bedrock and Azure Foundry agents integration in Agent Crew | Allows seamless import of external agents from AWS Bedrock and Azure Foundry into ZBrain Agent Crew for unified, cross-platform agent orchestration. | | | Inference-based fact storage for contextual continuity | Enables agents in the ZBrain Agent Crew to retain and reference prior conversation facts using Inference or Last Conversation memory modes. | | | AI-powered instruction generator in Agent Crew | Adds a “Generate” option to the Agent Crew instruction panel, enabling users to auto-generate or refine agent instructions using contextual input. | | | Optimized deep research output in agent crew | Enhances the Deep Research tool to deliver succinct, resolution-focused insights within Agent Crew. | | | Multi-document upload and management in the agent dashboard | Allows users to upload, view, and manage up to 10 documents simultaneously within the Agent Dashboard | | RAG | Editable chunk review before final knowledge base creation | Introduces an editable “Review Extracted Knowledge” step during KB setup, allowing users to refine extracted content before finalization. | | | Operator role management in the knowledge base | Extends Operator invite and access control functionality to the Knowledge Base module | ## New features ### Flows #### **Copy and paste workflow steps** ZBrain Builder enables users to copy any existing workflow step and paste it immediately after a selected step within the same workflow or across different workflows. This introduces a new capability that reduces repetitive configuration, allowing users to replicate the steps without manually rebuilding the logic while preserving full edit control over the duplicated step. **Navigation:**\ ZBrain Builder → Flows → Create New / Open Existing → ⋯ Menu (on a step) → Copy / Paste After
**Key outcomes:** * Accelerates workflow creation by eliminating the need to manually rebuild similar logic. * Enhances modularity and reusability across different automation Flows. * Minimizes user effort while maintaining consistency in the configuration and behavior of Flow pieces. * Supports scalable and agile workflow development, especially in large enterprise environments. #### **Version selection for individual workflow piece** ZBrain Builder introduces a version-selection dropdown for workflow pieces, allowing users to switch between available versions of any piece within a Flow. This upgrade supports more precise version control and easier rollbacks or upgrades during app, Flow, or agent iterations. After updating the version, all configuration settings for the component are reset to the default values of the newly selected version. **Navigation:**\ ZBrain Builder → Flows → Open Existing → Select Any Step (Piece) → Pieces Version dropdown (bottom of step configuration panel)
**Key outcomes:** * Facilitates rollback or testing of different logic versions with minimal configuration overhead. * Enhances stability by preserving upstream/downstream mappings during step-specific version changes. #### **Model update: Claude 4.5 Sonnet and GPT-5.1 support in the Ask AI Model piece** ZBrain Builder now supports Claude-4.5 Sonnet and GPT-5.1 in the Ask AI Model piece, enabling users to leverage the latest advancements in large language models for more sophisticated reasoning, summarization, and generation tasks within Flows. **Navigation:**\ ZBrain Builder → Flows → New Flow → ZBrain->Add/Select Ask AI Model piece → Model dropdown
**Key outcomes:** * Empowers users to build Flows with cutting-edge reasoning and generation capabilities using Claude-4.5 Sonnet or GPT-5.1. * Enhances model flexibility within workflows without affecting previous configurations. * Ensures traceability through explicit model labeling in run summaries. * Expands experimentation and performance tuning opportunities with newer LLMs. #### **New model support: Gemini-3-Pro-preview in the Ask AI Model piece** ZBrain now supports **Gemini-3-Pro-Preview** in the **Ask AI Model** step, enabling users to run Flows with one of the latest-generation Google models. This update expands the platform's model flexibility and enables more advanced reasoning and content generation capabilities. **Navigation:**\ ZBrain Builder → Flows → New Flow → ZBrain->Add/Select Ask AI Model piece → Model dropdown
**Key outcomes:** * Enables experimentation and deployment with Gemini’s latest LLM advancements. * Enhances performance, creativity, and comprehension for applicable Flows. * Maintains backward compatibility and configuration stability. #### **Document Intelligence and Multimodal Document Intelligence in Flow** ZBrain Builder now supports two new content extraction actions, Document Intelligence and Multimodal Document Intelligence, enabling users to ingest and process files directly from publicly accessible URLs. This capability improves content extraction flexibility across a wide range of formats. **Navigation:**\ ZBrain Builder → Flows → Create New / Update Existing → Search for **ZBrain Document Intelligence**
**Key outcomes:** * Automates document extraction from PDFs, images, and 15+ file formats, significantly reducing manual effort and turnaround time. * Ensures higher fidelity in capturing structured and unstructured data from complex documents through multimodal extraction. * Enables seamless integration of extracted content into downstream workflow steps, accelerating document-driven processes such as compliance and reporting. * Ingests diverse document types from public URLs without requiring format conversions or manual uploads. * Streamlines how teams incorporate AI into document workflows, improving productivity across departments such as legal, finance, HR, and procurement. ### **Agents** #### **Import external agents from AWS Bedrock into ZBrain Agent Crew** ZBrain Builder now enables users to import external agents created on **AWS Bedrock** into ZBrain’s Agent Crew, allowing seamless execution of cross-platform workflows within a unified AI environment. **Navigation:** ZBrain Builder→ Agents → Create New / Edit Existing → Create or Update Agent Crew → Add Agent → Import External Agent
**Key outcomes:** * Enables cross-platform orchestration, combining external Bedrock agents with ZBrain-native agents in unified workflows. * Offers enterprise flexibility by allowing organizations to reuse existing Bedrock agents inside ZBrain Builder without code migration. * Accelerates multi-agent application development by integrating third-party agents directly into the ZBrain ecosystem. * Simplifies operational scalability with centralized workflow design and execution across platforms. #### **Import external agents from Azure Foundry into ZBrain Agent Crew** ZBrain Builder now supports importing external agents created on Azure Foundry into the ZBrain Agent Crew, enabling users to orchestrate intelligent, cross-platform workflows within a unified agentic AI environment. **Navigation:** ZBrain Builder → Agents → Create New / Edit Existing → Create or Update Agent Crew → Add Agent → Import External Agent
**Key outcomes:** * Enables interoperable agent workflows by integrating Azure Foundry agents directly into ZBrain’s Agent Crew. * Enhances enterprise flexibility by supporting hybrid AI ecosystems across platforms. * Simplifies agent reuse and orchestration without code duplication or platform lock-in. * Provides clear configuration control with editable metadata, preserving ownership and flexibility within ZBrain Builder. #### **Inference-based fact storage for contextual follow-ups** ZBrain Builder now supports contextual continuity in Agent Crew interactions by allowing agents to store and reference facts from previous conversations. Users can enable either Inference Mode or Last Conversation Mode, under Crew or Tenant Memory, to ensure that follow-up questions are grounded in previously captured data. **Navigation:**\ ZBrain Builder → Agents → Create New → Create Agent Crew → Crew Overview-> Inference
**Key outcomes** * Enables agents to recall relevant facts from earlier interactions, ensuring more coherent and contextually aware follow-ups. * Supports nuanced back-and-forth exchanges without repeating inputs, delivering a more natural user experience. * Minimizes the need for users to re-explain information across queries—improving efficiency and engagement. * Enhances Crew collaboration by allowing multiple agents to access shared context, improving task coordination and output quality. #### **Agent Crew instruction panel – AI-powered instruction generator** ZBrain Builder now includes a ‘Generate’ option in the Agent Crew’s instruction panel, allowing users to auto-generate or refine instructions with contextual input. This feature simplifies instruction authoring by providing LLM-assisted guidance for more effective agent behavior. **Navigation:**\ ZBrain Builder → Agents → Create New→ Create Agent Crew → Instructions → Generate
**Key outcomes:** * Accelerates agent setup with intelligent instruction. * Reduces manual iteration by guiding users with well-structured prompts. ### RAG #### **Editable chunk review before final knowledge base creation** ZBrain Builder now allows users to review and edit extracted content before finalizing a knowledge base. This helps refine extracted data upfront, improving accuracy, relevance, and overall knowledge quality for downstream retrieval and reasoning. **Navigation:**\ ZBrain Builder→ Knowledge→ Create New → Review Extracted Knowledge
**Key outcomes:** * Empowers users to curate high-quality knowledge before ingestion into the knowledge base. * Prevents erroneous or irrelevant content from being embedded in the KB. * Increases transparency with source-level traceability post-deployment. * Enhances retrieval accuracy and downstream AI performance. #### **Operator role management in the knowledge base** To streamline permission control, the operator invite functionality is now available under the **Knowledge Base → User Management** section. This aligns access management capabilities in the Knowledge Base module, enabling administrators to grant or revoke Operator-level access directly. **Navigation:**\ ZBrain Builder → Knowledge→ Select KB Record → User Management
**Key outcomes:** * Expands administrative control by allowing KB-specific Operator permissions. ## Enhancements ### Flows #### **New vector store support: ChromaDB for knowledge base creation and retrieval** ZBrain Builder now supports ChromaDB for knowledge base creation, enabling users to build performant, lightweight retrieval systems and test them seamlessly within connected applications. **Navigation:**\ ZBrain Builder→ Knowledge Base → Create KB → Select Vector Store = Chroma
**Key outcomes:** * Simplifies the development of AI applications with integrated retrieval grounded in custom knowledge. * Enables rapid iteration and validation through native retrieval testing in connected apps. ### Agents #### **Optimized Deep Research output in Agent Crew** ZBrain Builder has enhanced the Deep Research tool within the Agent Crew framework to produce more concise and actionable results. This update significantly reduces unnecessary token consumption while maintaining the accuracy and utility of the research output. **Navigation:**\ ZBrain Builder→ Agents → Create New / Update Existing → Agent Crew → Add Deep Research Tool **Key outcomes:** * Reduces cost and improves efficiency in multi-agent or iterative workflows. * User receives distilled insights faster, without wading through excessive detail. * Requires no configuration change — just add the Deep Research tool to your Agent Crew and benefit from the new behavior instantly. #### **Multi-document upload** ZBrain Builder now supports uploading and managing multiple documents simultaneously in the Agent Dashboard, allowing users to compare, switch between, and process files without interrupting their workflow. This update streamlines instruction-driven processing and improves continuity across sessions. **Navigation:**\ ZBrain Builder→ Agents → Create New / Select Existing Agent → Agent Dashboard→Upload
**Key outcomes:** * Enables comparative analysis and multi-file processing without re-uploads. * Improves productivity by allowing seamless document switching without losing context. * Supports richer agent instructions tied to individual files for granular control.