What's new and changed in watsonx Orchestrate

IBM Software Hub Version 5.4.0

A new version of watsonx Orchestrate was released in June 2026 with IBM Software Hub 5.4.0.

Operand version: 8.0.0

This release includes the following changes:

New features

This release of watsonx Orchestrate includes the following features:

Handle errors in agentic workflows

You can add error handling to manage tool failures within your agentic workflows. When a tool fails, you can configure the workflow to automatically retry the operation. If the retry attempts fail, the workflow can display a custom error message to the user or continue along an alternative failure path based on your configuration. Now you can:

• Add error branches to any node in your workflow.
• Configure automatic retry logic with customizable retry attempts.
• Define alternative processing paths that execute when errors occur.
• Display custom error messages to users when retry attempts fail.
• Loop back to the original node after fixing issues.
• View error handling paths in flow inspector and observability traces.
• Maintain partial results when errors occur in multi-step processes.

Use this feature to build resilient workflows that can handle API failures, timeout scenarios, and data validation errors without losing progress or requiring manual intervention. For details, see Adding tools.

View flow run data on the canvas

You can open the flow inspector directly from the flow builder by clicking the Open flow inspector icon. Access workflow execution history and inspect how a workflow ran from start to finish without leaving the flow builder interface. Now you can:

• Click the Open flow inspector icon in the flow builder to access execution history.
• View the list of workflow runs with status, initiator, duration, and environment details.
• Filter workflow runs by time range, status, environment, agent, or initiator.
• Select a specific flow run to view detailed execution information.
• Review flow events showing the sequence of actions taken during the run.
• Inspect flow parameters including inputs, outputs, and flow variables.
• Open nodes directly in the flow builder from the flow run details.

Use this feature to quickly diagnose flow issues, review execution patterns, and understand workflow behavior directly from the flow builder interface. For details, see Inspecting agentic workflows.

Control builder access to embedded security settings

As an administrator you can now control whether builders can modify embedded security settings, ensuring only authorized personnel manage security configurations. By default, builders have access to configure security settings. Now you can:

• Toggle the Allow Builders to manage security settings option in the Embedded Security page.
• Restrict builders to read-only access of the Embedded Security page.
• Prevent builders from toggling the Security switch or modifying the public key field.
• Block API requests from builders to embedded security endpoints (returns 403 Forbidden errors).
• Maintain full administrator access to all security settings regardless of builder access setting.
• Apply changes immediately to all builders in your tenant.

Use this feature to enforce security governance policies and ensure only authorized personnel can modify embedded chat security settings. Changes take effect immediately and can be reversed at any time. For details, see Restricting builder access to security settings.

Work with time-based data in agentic workflows

You can handle time-based information in your agentic workflows with new Date/Time and Date/Time range fields, making it easier to schedule appointments, track work hours, and manage time-sensitive processes. Capture and process time data with time zone support. Now you can:

• Define time variables that store time values (hours and minutes with optional timezone).
• Use the date and time data type to store both date and time information in a single variable.
• Select from a comprehensive list of time zones, which is displayed with UTC offset and city name.
• Let users select time values by using intuitive time picker widgets in chat.
• Handle time ranges by defining start and end times for scheduling scenarios.

For details, see Date/Time, Date/Time range and Input, output, and variable types.

Control model selection visibility

You can control whether builders see model selection options in the UI. By default, model selection is hidden for new tenants, simplifying the builder experience for organizations that prefer to use default model configurations. Now you can:

• Turn model selection on or off from the Settings page.
• Hide model selection for agents while keeping it available for gen AI nodes in agentic workflows.
• Maintain existing model selection settings for current tenants.
• Enable model selection when builders need to choose specific models.

Use this feature to standardize on specific models while reducing complexity for builders who don't need to make model choices. For details, see Managing model selection settings.

Upload multiple files to tools

You can upload multiple files simultaneously to tools with configurable file type restrictions, size limits, and validation rules up to 30 MB per file. Now you can:

• Upload up to 100 files in a single operation.
• Configure maximum file size per file (up to 30 MB) and total upload size.
• Define allowed file types such as documents, spreadsheets, images, audio, and code files.
• Use the new WXOFile class for efficient file handling with lazy loading.
• Customize the file upload prompt message.

Customize tool fields for domain agents

You can customize tool fields for domain agents directly in the UI to accommodate client-specific field names and custom implementations without requiring code changes. Now you can:

• Add, remove, or modify tool field names during agent configuration.
• Adapt tools to match client-specific field naming conventions, for example, "ID" versus "IBMid".
• Configure tools for custom Salesforce, ServiceNow, and other integrations.
• Reduce tool errors caused by field name mismatches.

For details, see Customizing the schema for tools.

Use a schema from a list of predefined schemas

You can select from a list of predefined schemas when configuring structured document extractors in your agentic workflows. Simplify configuration by using ready-to-use schemas for common document types such as bank statements, invoices, and insurance claims. Now you can:

• Choose from a curated list of predefined schemas for common document types.
• Automatically add multiple fields associated with the selected document type.
• Reduce manual effort by avoiding the need to add fields one at a time.
• Quickly configure document extractors without manually defining field structures.

For details, see Use a schema from a list of predefined schemas.

Voice security for embedded chat

Voice capabilities in embedded chat now support both authenticated and anonymous modes. When security is enabled, voice audio streams use JWT-based authentication, binding each session to the user’s identity. For details, see Enabling voice capabilities in the embedded agent.

Updates

The following updates were introduced in this release:

Use gpt-oss-120b for your Agent development needs

You can use gpt-oss-120b model, an OpenAI’s open-weight model designed for

• Powerful reasoning and various developer use cases
• Answer conversational search questions
• Rewrite user questions to an understood format for conversational search
• Gather information to fill in variables in custom actions
• Select, connect, and coordinate multiple tools or APIs by using agentic AI
• Use prebuilt agentic AI agents that target specific domains

Agent Assist integrations and Dashboard updates

Preview: This feature is currently available only in preview.

With watsonx Orchestrate integrated into Genesys Cloud CX, you can interact seamlessly with both AI-powered and human-assisted support across voice and digital channels:

• Talk to AI agents over the phone: By using the Genesys Audio Connector integration, you can place or receive phone calls that are answered by AI agents. You can speak naturally, ask questions, and get real-time responses through voice conversations on the Genesys platform.
• Chat with AI agents through text: With the Genesys Bot Connector integration, you can have text-based conversations with AI agents. You can chat through web or digital clients to get quick answers, guidance, or support without making a phone call.
• Get more accurate help from human agents: Through the Agent Assist integration in the Genesys Agent Desktop, when you interact with a human agent, the agent is supported by AI. Agent Assist automatically surfaces relevant information from company documents such as standard operating procedures, product manuals, and knowledge bases, helping the agent answer your questions correctly. For details, see Learn more.

Get human help with full conversation context

In complex situations you might need human support, watsonx Orchestrate ensures a smooth and seamless transition to support on:

• Be transferred to a human agent during a voice call: If you are on a phone call with an AI agent and your issue needs human assistance, watsonx Orchestrate can transfer the call to a human agent using the Genesys Audio Connector. The human agent receives a summary of the conversation and the full transcript, so they already understand your issue and can continue the conversation without asking me to start over.
• Be transferred to a human agent during a chat: If you are chatting with an AI agent and need further help, watsonx Orchestrate can hand off the conversation to a human agent through the Genesys Bot Connector. The transfer includes a conversation summary and the full chat transcript, allowing the human agent to pick up right where the AI left off—without repetition.
• Get benefit from real-time transcription during voice escalation: After your phone call is transferred to a human agent, watsonx Orchestrate stays on the line silently and generates a real-time transcript of what you say. This helps the human agent clearly understand me, verify details if needed, and avoid any misunderstandings.

Get intelligent recommendations

watsonx Orchestrate enhances the user experience by equipping human agents with the right information at the right time. As a user, you can get the following agents:

• Company knowledge - Get accurate, policy‑aligned answers

  You receive responses based on company knowledge such as standard operating procedures, product manuals, and internal documents. This ensures your questions are handled correctly and consistently.

• Pre-transfer summarization - Move smoothly from AI to human agents

  If your call or chat is transferred, you don’t need to repeat yourself. watsonx Orchestrate creates a pre‑transfer summary of your conversation and shares it with the human agent, enabling a seamless handoff.

• Post-call summarization - Get better continuity after the interaction ends

  Once your call or chat is complete, watsonx Orchestrate generates a post‑call summary of the entire conversation. This summary is reviewed and added to your record, so future interactions start with full context.

Customer inquiries in Agent Assist dashboard

• Get faster, real-time help: You receive AI-assisted recommendations instantly as the conversation unfolds, helping address your queries more quickly.
• See your intent identified automatically: When you submit a new query, it appears in the Customer Inquiries window, where AI analyzes it in real time and identifies the most relevant intent.
• Act on recommended intents and content: Each intent includes prioritized actions and related knowledge or documents.
  • Click Run to execute an intent and view the response directly in the AI agent chat.
  • Click View to open supporting knowledge or documents in the AI agent chat without switching screens.

Benefit from prioritized guidance: Intents are ranked by priority set by the agent, so the most relevant actions and information are addressed first.

Use Transcript chat for voice calls

The Transcript chat is a real-time conversation display component within the Agent Assist Dashboard. The Transcript chat window is only available in voice calls for your use. You need an external audio adapter for voice call support.

Integrate watsonx Orchestrate with Genesys chat conversations by Chat connector

The newly added chat connector component enables watsonx Orchestrate to integrate with Genesys chat conversations. You can use the existing adapters as in previous on-prem environment.

New prebuilt agents

Category	Agent name
IT	Knowledge Manager in ServiceNow
Procurement	Invoice Manager with Ariba Supplier Invoices with Ariba
Legal	Legal Operations Manager Legal Service Requests Legal Matter Management

For the complete list of available prebuilt agents and tools, see List of prebuilt agents.

New enhancements

Enhancement	Description	Affected role
Voice configuration in deployment summary	View voice configuration details in the pre-deployment summary to confirm settings before deployment.	Builder or Writer
Improved connections configuration	Enhanced visual consistency, clearer OAuth flows, and better error handling in the connections manager.	Builder or Writer
Enhanced deployment error handling	Nondismissible error notifications for blocking issues and clearer deployment status indicators.	Builder or Writer
Improved procurement agents	Procurement agents that are migrated to GPT OSS models for better performance and response quality.	User or WO User
Simplified mini pallet interface in flow builder	The mini-pallet features a cleaner, more intuitive design with visual improvements that reduce complexity while maintaining all existing functions. When adding tools, you are guided to switch to the main pallet to drag tools onto the canvas.	Builder or Writer
Improved /runs API performance	The /runs API endpoint is faster and more responsive, improving reliability when you run assistants and orchestration workflows, especially during peak usage.	Builder or Writer
Flexible model selection for document processing	Document extractor (structured and unstructured) and document classifier nodes list the models available to you by default. If the model you need is not shown, you can manually enter a custom model from AI Gateway.	Builder or Writer

New features from 5.3.1 patches

This release of watsonx.governance™ includes the following features that were introduced in IBM Software Hub Version 5.3.1 patches:

Protect sensitive data with PII masking

You can configure Personally Identifiable Information (PII) masking to protect sensitive information in observability traces. Choose between the selective masking by using automatic detection or full masking of all trace content.

You can:

• Enable PII masking at the tenant level from Settings
• Select the selective masking to automatically mask common PII patterns (emails, phone numbers, SSNs)
• Select full masking to remove all user input and agent output from traces
• View clear warnings about the impact on testing and monitoring
• Maintain historical traces (masking applies only to new traces)

For details, see PII masking.

Inspect workflow runs with flow inspector

You can inspect agentic workflow runs by using flow inspector. View run status, initiator, agent, start time, environment, and instance ID. Search for specific runs, filter by multiple criteria, review workflow events and workflow parameters, view AI-generated summaries, and open workflows in flow builder to update configurations.

Now, you can use flow inspector to identify run patterns, trace execution paths, diagnose errors, and validate workflow behavior across different run scenarios. For details, see Inspecting agentic workflows.

Interact effectively with chat with improved chat accessibility

Chat now meets European Accessibility standards (RGAA) with comprehensive keyboard navigation and screen reader support.

You can:

• Chat with a keyboard: Delete threads, minimize full‑screen view, and download files without needing a mouse.
• Stay informed in real time with screen readers: Incoming messages and important UI changes are clearly announced.
• Understand and control what you see: Show or Hide Reasoning uses proper buttons that work consistently with assistive technologies.
• Get started smoothly: The welcome prompt carousel is readable and announced correctly by screen readers.
• Move confidently through chat: Improved focus handling keeps keyboard and screen‑reader users oriented at all times.

These accessibility improvements can ensure that all users can effectively interact with chat, regardless of how they access the interface.

Preview - Assign users to document processing reviews

You can assign specific users to validate document processing when user intervention is required. Configure document extractor and document classifier nodes to route validation tasks to designated users.

You can:

• Configure user reviews to trigger based on extraction confidence thresholds or empty values for document extractors
• Select specific fields that require review in document extractors
• Route classification reviews when documents are classified as "Other" for document classifiers
• Receive chat notifications when reviews are required

When a document requires review, the assigned user receives a notification in chat. They can validate extracted values or classification results, make necessary changes, and submit the review. The workflow continues with the updated values. For details, see Configuring user reviews for document extractors and Configuring user reviews for document classifiers.

Add custom action buttons to forms

You can add more action buttons to forms to support complex workflows like scheduling, approvals, and multi-step processes. This feature extends beyond the default Submit and Cancel buttons.

You can:

• Add multiple more buttons with configurable labels and actions
• Define whether each button submits or cancels the form
• Display up to three buttons directly, with more buttons in an actions list
• Use more action buttons in both chat and Flow builder

The more action buttons allow you to create sophisticated form-based workflows. For details, see Adding interactions.

Access context variables in flow builder

Retrieve and update shared workflow data by using system.context.get_variable() and system.context.set_variable(). System context variables persist for the entire duration of a workflow run, so values can be reused across multiple steps.

You can also access built-in variables, such as wxo_email_id, wxo_tenant_id, and wxo_user_name at any point in your flow without setting them manually. For details, see System context variables

Configure user review for document processing nodes

Get notified in chat when a document processing manual review activity is assigned to you in a flow.

This can ensure that you respond to review requests promptly. Notifications appear directly in your chat thread, so you can complete reviews without switching tools. For details, see Configuring a user review for document extractors and Configuring a user review for document classifiers.

Manage versions for agents and tools

Use built-in version management for agents and tools in watsonx Orchestrate. Version management helps you review changes, choose specific versions, and reuse earlier configurations, so you can build safely without disrupting users.

You can:

• View up to five recent versions with summaries and change logs
• Switch versions from View recent versions in the catalog
• Add a specific tool or agent version as a collaborator

Create a new agent from an older version using Use as template

Add agent nodes to your flows

Add agent nodes to agentic workflows in flow builder. Agent nodes enable you to build agent patterns by combining agents with workflow constructs such as foreach loops, branches, and code blocks.

You can:

• Add any agent from your local instance as a workflow node
• Configure the agent node with a user prompt and workflow variables
• Request additional information from the workflow at run time when needed

This feature enables you to build predictable solutions by routing tasks to specialized agents within structured workflows.

User silence management for Voice AI agents

Configure how your Voice AI agents handle user silence during phone conversations. This helps you create more natural interactions, reduce dropped calls, and avoid unnecessary wait time. You can:

• Set a Silence threshold after which the agent identifies the user as silent
• Define a Check-in message that is played when silence is detected. The agent repeats its last response before playing this message
• Configure Number of check-in repetitions to prompt the user before ending the call
• Specify a Call ending message to play before the call ends when the maximum number of attempts is reached.

These controls help you keep conversations moving, give callers thinking time, and prevent calls from ending too early or staying open indefinitely

Customize your embedded chat header

Add custom buttons and dropdown menus to the embedded chat header. With the new APIs, you can create programmatic customization of embedded chat header items in the full-screen form factor.

APIs included:

• updateCustomHeaderItems()
• getCustomHeaderItems()
• getLocale()

You can:

• Place controls on the left or right side of the header
• Create dropdown menus up to two levels deep
• Update labels, icons, and actions at run time

This lets you embed branded controls, shortcuts, or navigation directly in the chat header. For example, add quick links to help documentation, user settings, or context-specific actions. For details, see Details of instance methods.

Preview - Handle multiple activities in chat threads

Handle multiple activities between chat threads by picking the activities that you want to work on from an activity queue.

You can:

• See the activities that are pending in your thread
• Open and switch between queued activities
• Complete tasks directly in chat
• Return control to the agent automatically after each task

When a workflow includes multiple approval steps or user reviews, the activity queue consolidates all actions in a single, and organized view. This approach minimizes context switching and helps keep the workflow moving efficiently.

The notification and queue system uses a session‑based design, helping can ensure that alerts remain focused on current activity and do not persist across page reloads. For details, see Designing high-performance AI agents.

Use multiple authentication types for ServiceNow tools and agents

Choose from multiple authentication methods when configuring out-of-the-box ServiceNow tools and agents to align with your organization's security requirements.

Supported authentication types:

• OAuth 2.0 Authorization Code
• OAuth 2.0 Password Flow
• OAuth On Behalf Of (OBO)
• OAuth 2.0 Token Exchange
• Bearer Token

You see only the authentication types that are supported by each ServiceNow tool or agent, making it easier to align with your security policies and use your existing enterprise identity infrastructure. For details, see Connecting to ServiceNow.

Configure embedded chat security at the tenant level

Controls security for all embedded chat implementations with a single tenant-wide toggle.

You can:

• Enable or disable embed security with one click
• Upload your client public key
• Generate an IBM-provided key to encrypt data and protect specific sensitive information
• View keys clearly in the UI with save and discard controls
• Receive warning prompts before disabling security

This can ensure consistent, tenant-wide protection for all embedded chat implementations while simplifying key management.

Multilingual support for welcome messages and starter prompts

Create welcome messages and starter prompts once in English. The system automatically converts them when users change their language setting.

You can:

• Create content once and reuse it globally
• Support multiple languages without extra configuration
• Provide content instantly in users' preferred languages
• Maintain one source of truth for your content

This works in both the Orchestrate Chat and embedded chat, reducing maintenance while improving accessibility for global teams. For details, see Customizing the welcome message and starter prompts.

Issues fixed in this release

The following issues were fixed in this release:

Customer-reported issues fixed in this release

For a list of customer-reported issues that were fixed in this release, see the Fix List for IBM Cloud Pak for Data on the IBM Support website.

Security issues fixed in this release

The following security issues were fixed in this release:

CVE-2026-5201, CVE-2026-20608, CVE-2026-20635, CVE-2026-20636, CVE-2026-20643, CVE-2026-20644, CVE-2026-20652, CVE-2026-20664, CVE-2026-20665, CVE-2026-20676, CVE-2026-20691, CVE-2026-21226, CVE-2026-21441, CVE-2026-22795, CVE-2026-22796, CVE-2026-23490, CVE-2026-23865, CVE-2026-23868, CVE-2026-24051, CVE-2026-25518, CVE-2026-28857, CVE-2026-28859, CVE-2026-28871, CVE-2026-31826, CVE-2026-32285, CVE-2026-32287, CVE-2026-33123, CVE-2026-33230, CVE-2026-33231, CVE-2026-33236, CVE-2026-33699, CVE-2026-33815, CVE-2026-33816, CVE-2026-40179, CVE-2026-41305

CVE-2025-1194, CVE-2025-1386, CVE-2025-2099, CVE-2025-3263, CVE-2025-3264, CVE-2025-3933, CVE-2025-5197, CVE-2025-6051, CVE-2025-6638, CVE-2025-6921, CVE-2025-9086, CVE-2025-11065, CVE-2025-11187, CVE-2025-12084, CVE-2025-13601, CVE-2025-13836, CVE-2025-14104, CVE-2025-15467, CVE-2025-15468, CVE-2025-15469, CVE-2025-22868, CVE-2025-22869, CVE-2025-22870, CVE-2025-22872, CVE-2025-29786, CVE-2025-30204, CVE-2025-31133, CVE-2025-43213, CVE-2025-43214, CVE-2025-43457, CVE-2025-43511, CVE-2025-45582, CVE-2025-46299, CVE-2025-47906, CVE-2025-47907, CVE-2025-47911, CVE-2025-47912, CVE-2025-52565, CVE-2025-52881, CVE-2025-58185, CVE-2025-58187, CVE-2025-58188, CVE-2025-58189, CVE-2025-61723, CVE-2025-61724, CVE-2025-61726, CVE-2025-61727, CVE-2025-61728, CVE-2025-61729, CVE-2025-61730, CVE-2025-61731, CVE-2025-63811, CVE-2025-66199, CVE-2025-66418, CVE-2025-66471, CVE-2025-68121, CVE-2025-68156, CVE-2025-68160, CVE-2025-68973, CVE-2025-69418, CVE-2025-69419, CVE-2025-69420, CVE-2025-69421

CVE-2024-45337, CVE-2024-45338

Deprecated models

The following models are deprecated in this release:

Deprecated models remain functional until their removal date but may display removal notifications in the UI. Switch to supported models before removal to avoid disruption. Recommended model is openai/gpt-oss-120b.

• ibm/ibm-granite-8b-unified-api-model-v2
• ibm/granite-3-8b-instruct
• meta-llama/llama-3-1-70b-instruct
• meta-llama/llama-3-2-90b-vision-instruct