Known issues and limitations

Get a quick overview of the known issues and limitations for IBM watsonx Orchestrate.

watsonx Orchestrate agentic experience

Area affected Description
Importing MCP toolkit using ADK Leaving the watsonx Orchestrate ADK setup idle for too long can cause an error "Client.Timeout exceeded while awaiting headers. 500 Internal Server Error" when importing an MCP toolkit. To resolve the issue, run "orchestrate stop" and then "orchestrate start".
watsonx Orchestrate preview chat The option to enter team credentials for an app or tool isn’t available in the preview chat for builders. As a result, you might see an error: "Authentication failed using the team credentials. Please contact your administrator to update the credentials."
Uploading files on watsonx Orchestrate chat When uploading files through the watsonx Orchestrate chat, trying to re-upload an existing file will cause an error: "File is already uploaded". To resolve the issue,  ensure you're uploading unique files each time.
The Catalog does not have prebuilt agents and tools The Catalog might not have prebuilt agents and tools available after the instance provisioning. If you didn’t request access to prebuilt agents and tools, complete the following form: Preview access request form

After approval, the agents and tools for productivity, HR, procurement, and sales will be available in a few days. In the meantime, you can explore the following options within the product to continue your activities smoothly:

  • Explore the list of agents and tools: Get more information on the use cases that the agents for productivity, HR, procurement and sales can cover. For more information, see List of prebuilt agents.
  • Create your own tools: Use the IBM watsonx Orchestrate Agent Development Kit to easily create tools programmatically. For more information, see Creating your first agent on agent builder.
  • Build your own agents: Use the agent builder to build custom agents based on your business knowledge, tools or other agents that you add for multi-agent orchestration. For more information, see Creating agents.
External agents Step loader is not visible when two external agents run in the same thread within the native agent. The thread remains with the external agent.
Agent tools Tool execution fails on the first attempt to invoke the tool and succeeds during the second attempt on the same thread. If you encounter a failure after a prolonged period of inactivity, retry the execution once.
Agent tools An agent fails to deploy if it includes a flow that was created previously in the technology preview. To resolve this issue, open the flow in the flow builder and edit the name, or description and save it. Saving the flow updates the flow configuration to work with the general availability release.
LLM in agents The large language model (LLM) fails to interpret the successful outcome of a tool execution. You might get LLM responses like "I am unable to identify a tool to execute." and "I have encountered an error calling the tool. Please try again".
Prompting in agents In the watsonx Orchestrate chat of new experience, if you prompt multiple questions to the agent continuously, the agent does not invoke any tool and responds with None.
Interactions containing media and UI widgets are not supported Interactions with AI assistants are limited only to text. Response types such as media and UI widgets are not supported.
AI assistants sourced from the same instance as a collaborator agent AI assistants from the AI assistant builder must be sourced from instances other than the instance that you use to build the agent. It is not supported adding AI assistants sourced from the same instance as a collaborator agent.
AI assistants hosted on IBM watsonx Orchestrate on AWS Adding AI assistants from the AI assistant builder as a collaborator agent is available only for IBM watsonx Orchestrate on IBM Cloud. It is not supported for IBM watsonx Orchestrate on AWS.
Interacting with different tenants in the same browser window is not supported When users attempt to access multiple tenants of watsonx Orchestrate from the same browser window, session cookies cause tenant data to be shared across active sessions. As a workaround, use separate browser profiles, different browsers, or Firefox containers to isolate tenant sessions and prevent shared cookies from causing conflicts.
Inconsistent transfer between multi-agent collaborators In a multi-agent setup, the collaborator agent does not return control to the supervisor agent after completing its task. As a workaround, update the collaborator agent’s configuration to ensure it transfers control back to the supervisor agent after completing its task.
Steps to update Instructions:
  • Open the collaborator agent in the Agent Builder
  • Go to Behaviors > Instructions.
  • Add a clear instruction specifying that the agent must transfer control back to the supervisor agent after completing its task.
    Note: This workaround does not apply when the collaborator agent is an external agent.
    		•
    Document upload fails The document uploaded in the AI agent chat might fail for files exceeding 8MB due to a timeout limit. To avoid this issue, ensure that the uploaded documents are 8MB or smaller.
    Repetitive responses by model Llama 3.2 70B Instruct The model might exhibit repetitive patterns, especially when engaging in long conversations or certain scenarios, generating the same response regardless of changes in the input.
    Llama 3.2 70B Instruct model context length The model's capacity to retain past conversation history might decrease based on available memory, impacting the maximum supported context length.
    Chat with documents is not supported in voice (Preview) The AI agent's document chat functionality is unavailable when using voice input.
    Embedded chat disappears When you are using the embedded full screen chat layout with the left sidebar showing chat history, minimizing and reopening the chat takes you to a new chat instead of the one you were using. Attempting to access the previous chat from the sidebar causes the embedded chat to disappear.
    AI agent chat fails in the AWS Mumbai region AI agent chat shows an error in the AWS Mumbai region if the tenant has no agents, because the system links the default agent to the wrong model. To fix this, create an agent before using AI agent chat.

    Building AI assistants

    Area affected Description
    Importing watsonx Assistants Due to a temporary technical glitch, the unpublished watsonx Assistants are displayed in the dropdown menu during the import process.
    Digression Users can't digress to the default action behavior when they interact with actions from skills on AI assistants.
    Language support for AI assistant builder The AI assistant builder supports the English language only.
    Authentication on AI assistant API with IP address restriction To use AI assistant builder API with an IP address restriction, use a bearer token in an authorization header instead of an API key.

    Generic issues

    Area affected Description
    Issues with watsonx Orchestrate login experience on Chrome v128 When logging in to watsonx Orchestrate using Chrome version 128.0.6613.85, the landing page is either slow or doesn't render. A blank page appears upon selecting a tenant. Browser refresh may not help, but holding the shift key while refreshing the browser takes you to the tenant. To resolve the issue, upgrade Chrome to version 128.0.6613.120 (Official Build).
    Application tiles load slowly on the home screen The loading and display time of application tiles on the watsonx Orchestrate home page may be prolonged, particularly when there are many applications present on your home screen. The issue will be prioritized and considered as part of future enhancements.
    Log in to watsonx Orchestrate Sometimes, when you log in to watsonx Orchestrate, it throws an error that says "We couldn't find you", which results from cache issues. To troubleshoot this problem, clear your cache.
    Memory persistence in watsonx Orchestrate When you refresh the browser, or log out of your active watsonx Orchestrate instance, watsonx Orchestrate is reset. For example, if you use the Find candidates for a job skill to generate a list of candidates then click the header to go back to the home page, your results from the previous query will be gone. To avoid this limitation, consider completing the course of action before leaving the context of your actions (downloading and emailing applicants after retrieving the candidate list).
    Authentication on watsonx Orchestrate API with IP address restriction To use the watsonx Orchestrate API with an IP address restriction for watsonx Orchestrate on IBM Cloud, use a bearer token in an authorization header instead of an API key. For more information about IP address restriction on IBM Cloud, see Allowing specific IP addresses for an account.

    IBM SaaS Console

    Area affected Description
    Billed usage The IBM SaaS Console is not showing billed usage information.

    watsonx Orchestrate API

    You might encounter known issues and limitations when you use the methods of the watsonx Orchestrate API.

    To get a quick overview of the known issues and limitations for the API, see the IBM watsonx Orchestrate API documentation page.

    watsonx Orchestrate skill-based experience

    Following are the list of known issues and limitations for the skill-based experience in IBM watsonx Orchestrate.

    Skill-based actions

    The following table describes the known issues and limitations that are specific to skill-based actions in AI assistants:

    Area affected Description
    Bot fails with the error 'ERR_WO_001G' : 'str' object has no attribute 'get' Running a skill on the AI assistant builder with a newly created bot triggers the error message 'ERR_WO_001G' : 'str' object has no attribute 'get'. To resolve the issue, you must switch to the legacy chat and ensure that the bot environments are successfully created, then return to run the skill.
    Note: To switch from AI chat to the legacy chat, see Switching the chat experience.
    Skill execution fails with an internal server error for OAuth2 authentication When you run a skill and the token expires, the skill execution fails with an internal server error. To resolve the issue, retry the skill to refresh the token.
    Skill execution triggers a 504 Gateway Timeout error When you input a large volume of data through a skill, the time to process the data might exceed the predefined time limit set in the watsonx Orchestrate proxy, which results in a timeout error. As an alternative, you can use asynchronous skill execution.
    Setting credentials to skill-based actions Skill-based actions support only shared credentials that are set through Team credentials as credential type.
    Forms are not supported for third-party channels and custom integrations Third-party channels and custom integrations through the API don't support form-based interaction. As a workaround, update your custom skills to render their inputs and outputs as a multi-turn conversation. For more information, see Defining how to interact with a skill in a conversation.
    Multi-turn conversations are not supported for skill flows and most of the prebuilt apps and skills skill flows and most of the prebuilt apps and skills do not support multi-turn conversations and render skills' inputs and outputs as form-based.
    skill flows render as form-based regardless of their component skills All skill flows render as form-based regardless of whether their component skills are supported in multi-turn conversations when they are called independently.
    Actions based on asynchronous skills and long-running synchronous skills work on the AI chat only Actions based on asynchronous skills and long-running synchronous skills render inputs and outputs as a multi-turn conversation in the AI chat only.

    For more information about how inputs and outputs are rendered in skill-based actions, see the criteria at Defining how to interact with a skill in a conversation.
    For more information about the AI chat, see Using the AI chat.
    Slot filling is not available for skills that render as form-based Skills that render as form-based do not benefit from the slot-filling capability.
    Variable passing limitation Known the issues and limitations that are related to the variable passing:
    1. You can pass variables only for a new skill-based action. This feature does not work with skill-based actions that were created before the implementation of this feature.
    2. The output from the skills that have dynamic output forms are not correctly returned to watsonx Assistant.
    3. Updates to a skill are not supported for variable passing. If the schema of a skill or skill flow is updated on watsonx Orchestrate, the action that is calling this skill keeps pointing to the older version of the skill or skill flow.
    4. You cannot upload files to skill-based actions despite the skills having the file upload feature. Variable passing does not support files.
    5. You cannot pass variables to the input filters of skill-based actions.
    Actions disappear in the AI assistant builder     		When the builder returns from another page to the AI assistant builder, the actions disappear. Select the Actions tab again or refresh the page to solve the problem.
    Duplicate skills are created Duplicate skills are created due to a mismatch in the skill ID generation. Previously, API Management (APIM) generated skill IDs by converting space in the application title to '-', resulting in IDs like send---mail for the application title Send - mail. However, the backend now uses the skill server, which removes spaces in the application title and generates IDs like send-mail. These auto-generated IDs create duplicate entries for the same skill.
    Using form skills in the AI assistant builder You can use form skills only with web chat integrations.
    Enhancing the skill content in other languages You can enhance the skill name, description, input, and output parameters in the supported language. However, the skill and its parameters display in the default language in AI chat and AI agent configuration - Apps and Skills pages.
    Note: Actions based on custom skills automatic fallback to forms over multi-turn conversations when any of the criteria that are described at Defining how to interact with a skill in a conversation are met.

    Actions from web-based automation flows with UIAgent

    The following table describes the known issues and limitations that are specific to actions from web-based automation flows with UIAgent in AI assistants:

    Area affected Description
    Using the Select action when you create a new step for a UI flow The Select action supports only select or radio as attribute values of type for HTML elements.
    Use of the extension in multiple windows You can use the extension only one window at a time. If you try to open the extension panel in two or more windows, the extension does not work as expected and shows a blank screen.
    Typing and pasting on the same field If users try to type in a field, and later try pasting a value on the same field, the automation breaks.

    Building projects

    Area affected Description
    Building projects If you export a project, you should not rename the generated compressed file and try to import the compressed file again.
    Modeling data Do not create two data types where the only difference in their name is the capitalization of the first letter.
    Modeling data When there is an error in your data model, the error is not shown in the workflows in the project. Only the last working data model is shown in the workflows. You must fix the error in the data model so that you can see the latest data model in the workflows.
    Starting skills from a workflow Learn about the limitations when calling skills from a workflow and when using OpenAPI constructs.
    Starting skills from a workflow that uses filter input If a skill has a filter input, you must use the following syntax to specify the input: {"where":{"or":[{"id":1},{"id":2},...,{"id":20"},{"id":21}]}}, where id is the queryable parameter as documented in the Filter input for skills topic.
    Starting skills from workflow All skills that are started by a workflow must exist in the workflow originator’s skill set, and the application must be configured. Otherwise, the workflow fails. To resolve, add and configure the missing skills. For in-flight processes, you can ask an administrator to retry the skill activity by using the process inspector. Otherwise, restart the workflow after you add the skills.
    Workflow project For any watsonx Orchestrate tenants provisioned before 15 December that have the Skill studio feature, all the watsonx Orchestrate users must be synchronized to the Skill studio by the owner of the tenant or a new Admin that was added to the tenant. If the user synchronization is not done, the workflow project in the Skill studio doesn't work.

    The tenant owner or the new Admin needs to do the following steps:

    1. Log in to the watsonx Orchestrate tenant.
    2. From the main menu, click Project.
    3. In the Projects page, click the Workflow Admin button.
    4. In the Workflow Admin page, from the upper left menu, click User Management.
    5. Click User Synchronization.
    6. Click the Synchronize Tenant Users button.
    Removing and re-adding members When you remove a member from a tenant, the projects where that member is the only collaborator will be deleted. Before you remove a member, make sure to add other collaborators to those projects.

    When a member is removed from the team and re-added with the Builder or Admin role, the member's workflow permissions might be incorrect. This can cause errors when editing workflows that the member is not recognized as Author (Repository Write Access) or Administrator. To resolve this limitation, an Admin needs to complete the following:

    1. Go to the Manage Team page.
    2. From the Member tab, switch the role of the affected member to User, then back to the desired role. This updates the permission properly and prevents the "CWTBG0632E: You are not authorized" error.
    Creating workflow variables for starting projects When building a workflow project, you can create an activity to call a project. If you try to create variables for the project parameters from the Variables tab, you might not see the data types for the parameters of the activity. To resolve this, go to the diagram, click the project that you're calling, and click Define data mapping. The data types should then appear in the Variables tab.
    Publishing a workflow as a skill When you publish a new version of a workflow that contains enum inputs, the form displayed in the chat might show duplicate values in the drop-down.
    Previewing workflow projects When you preview a workflow that calls a separate project, you won't see any tasks related to that separate project in the preview mode task list. In order to work with the tasks from that separate project, you must go to the Tasks page, rather than the tasks page in preview mode.
    IBM Robotic Process Automation skills The IBM RPA skills are not working in the workflow, even though they are working in the watsonx Orchestrate chat.
    Importing samples and tutorials Only one user in each tenant can import each industry sample or discovery tutorial. The user who imported the sample or tutorial must add other users as collaborators for them to use the sample or tutorial.
    Publishing a project as a skill If your workflow project uses user type inputs or outputs, or contains nested user type inputs or outputs, you can’t publish it as a skill in the skill catalog. However, you can still call the project through another project.

    Publishing a project might fail when the following actions are taken:
    1. Create a project that includes a decision based on an imported project, and publish the project before 2024.
    2. Delete the project.
    3. Create another project that includes a decision based on the same imported project in 2024 or later.
    4. Publish the newly created project. This publication might fail.
    To publish the new project successfully, you need to remove the skill that was published before 2024 from the skill catalog first.
    Data mappings are not automatically updated for decision and skill activities in a workflow. When you update the inputs or outputs in the data mapping for a decision or skill implementation, the change is not automatically updated in the workflow activity, and you might still see the old types and parameters. Click Refresh connection in the workflow activity properties to reflect your changed inputs and outputs.
    Changing a role Changing your role from admin to builder or from builder to admin deletes all projects from your Projects page.
    Previewing or calling a workflow as a skill When you preview your workflow or call a workflow as a skill, you might see the error message “401 Worker object not found”.
    Changing input or output variables Users need to remove any previously published skills if there are any changes to the input or output variables from or to the document type.
    Language support for Skill studio The Skill studio supports the English language only.
    Expressions in workflows You need to specify an integer value or an integer variable when an integer is expected. For example, when you use the expression "the element at position <position_value> in <list_name>", you need to use an integer value or an integer variable for <position_value>. If you specify a number value or a number variable instead, your workflow will get the following error when it tries to execute the expression during runtime: "EL1074E: An exception occurred while compiling an exception".
    Returning numbers as outputs in decisions Users might see an error when they try to run a decision skill that returns a number as the output. This scenario occurs in the chat of the AI assistant builder.
    Skills with objects without attribute definitions If a skill parameter or nested skill parameter has an object attribute that has no attributes, it is ignored when you find and try to use it in a workflow skill activity. The skill activity does not show the object in its data mapping.
    Deleting components If you delete a component such as a workflow, decision, or generative AI, you might see the following issues or limitations:
    • If the deleted component was published, you must manually delete it from the skill catalog. To delete it from the skill catalog, go to Skill studio and delete the skill.
    • If the deleted component was published in a previous version of the project containing it, other projects that use the component might not work anymore. You must manually replace or remove the deleted component from other components and skills that use it. Broken references to the deleted component are not displayed while you are editing your project. You will get an error only when you try to test or use the component.
    • If you shared the deleted component previously, it will not automatically be deleted for other collaborators in this project. Share your changes after the deletion, so that other collaborators get the deletion update.
    Using multiple tabs in your browser When you use multiple tabs in your browser, for example to do some comparison or validation of your skill, you must refresh the tab to see updates that you did in other tabs.
    Skills with dynamic inputs for project workflows Skills that have dynamic inputs are not supported for projects workflows.
    Creating a workflow with an assignment activity that uses expressions When you create a workflow, add an assignment activity with expressions, and then invoke the workflow skill in the chat, a skill issue error (500) occurs.
    Publishing a ruleflow with functions When you toggle a ruleflow from private to public, all functions that exist in the ruleflow at that time are marked as public, and displayed in Skills and apps when the ruleflow is published.

    If you add new functions after toggling the ruleflow to public, these functions remain private until you toggle again to public.
    Cannot preview a workflow that contains the "Utility bill extractor" skill You might get a "401 - Unauthorized" error message when trying to preview a workflow that contains a "Utility bill extractor" skill, or if you start that skill from the chat. To prevent this, open Skill studio, go to Skills and apps > Skills > Configure prebuilt skills, and set the server URL of your app endpoint.
    Previewing a workflow that includes a generative AI activity The following exception might be thrown when you preview a workflow that includes a generative AI activity:
    Error code: model_unavailable
    Error message: Model <model_name> is currently unavailable, please try later.
    This exception might occur when there is not enough resource in watsonx.ai to call the generative AI skill many times. You might want to choose a different model or reduce the number of tokens, and try again.
    Calling skills from interactions When calling skills from interactions, your skill might not work if the inputs or outputs are not in the expected format. Your skill might not work if it fulfills one or more of the following conditions:
    • Your skill inputs are not wrapped.
    • Your skill has exactly one output value that is wrapped.
    • Your skill openAPI input and output property names are not in the expected format. The name must start with a letter, dollar sign, or underscore, and can only contain letters, numbers, dollar sign and underscore.
    To check whether your inputs or outputs are being wrapped:
    1. Go to the skill call "Select a skill" page.
    2. Right-click the webpage and select Inspect.
    3. Go to the network tab, and start a network log recording.
    4. On the webpage, select the skill that you want to use.
    5. In the developer console, select the entry with a name similar to skill?skillId=<skillname>, go to the Preview tab, open the component schemas subsection, check the requests and responses and look for the x-ibm-parameter-wrapper attribute. If the value is true, the value is being wrapped.
    Changing the operation ID after the skill is published If you change the operation ID of a skill after the skill has been published, you need to go to the Skills & Apps page and manually delete the previously published skill. Check the Added on field to make sure that you delete the correct version of the skill.
    Renaming your generative AI skill name or operation ID After creating your generative AI skill, you cannot change the skill name or the operation ID.
    Cannot open sample documents in an old project Sample documents are available for 90 days. If they expire, you need to re-upload the documents. This doesn't impact the document extractor skill that you trained with those sample documents after you publish the skill.
    Document extractor skills in workflows
    • Uploaded file names cannot have spaces. When you upload files to your document extractors, spaces in the file names might cause errors when previewing the workflow.
    • To publish from a workflow, click Publish twice in a row.
    • If you have multiple document extractors in a project, make sure the fields you add have unique names across all extractors, otherwise you might see an error saying Ambiguous in the review fields of your extractor in the workflow.
    • You might see an error when clicking Refresh connection, Define data mapping, or Set extraction rules from the workflow activities properties. To avoid this error, do either of the following:
      • Run the preview, then try to click those options again.
      • Remove and re-add the document extractor activity in your workflow.
    Calling skills with document parameters When calling a skill with a document parameter from a user interface interaction, use the Binary file type, even if the type of file is simple text. Calling the same skill directly in the workflow can allow a different setting.

    Building skills

    Area affected Description
    Mapping values from fields containing spaces or special characters watsonx Orchestrate fails to map values from fields containing spaces or special characters in a JSON path for skill flows, especially in nested fields. To resolve this, consider renaming the values to exclude special characters and spaces.
    Building skills If the summary and description fields in an OpenAPI specification are at the path level, then three skills are displayed for a single operation specification when you build skills by using an OpenAPI specification. As a workaround, remove the summary and description fields from the paths section to make the import succeed. For example, remove the summary and description fields as displayed under the path section: 
    ---------------
    "paths" : {
      "/Test_WO" : {
        "summary" : "Test_WO",
        "description" : "Testing",
    ---------------
    Building skills When you build skills through OpenAPI specifications, descriptions, default values, and enum values can't exceed the character limit of 64 characters.
    Building skills The requestBody property is not supported for OpenAPI skills based on the GET method.
    Building skills from RPA The output of skills that use RPA bots might not show values from output variables as a result. Sometimes, creating a new output variable on the bot and updating the skill might resolve the problem.
    Combining variables and default values in custom forms You can't combine variables from data mapping and default values in a customized input form.
    Logo of the third-party application If you notice a color variation in the application icon, you must revise the OpenAPI specification of the skill to avoid the usage of <style> tag within the <svg>. Refer Provide an icon to your application to view the format to provide the icon.
    Mapping inputs from phrases "patterns" are not currently supported in the x-ibm-nl-entities property, but will be considered as part of future enhancements.
    Configuring an imported app endpoint The "Test your connection" option in the Enhance this skill page doesn't use the custom endpoint to test the connection. Therefore, you can test the connection for your new endpoint with this feature.
    Configuring an imported app endpoint Even though it is necessary to reconfigure the app credentials to use its skills, the app appears as connected on the skill catalog if the app has a previous connection.
    Language support for OpenAPI builder The OpenAPI builder feature supports the English language only.
    Input fields don’t appear as defined in the x-ibm-order property The order of the input fields in the skill form might not appear as defined in the x-ibm-order property. You might encounter the following limitations when you use the x-ibm-order property in a skill specification:

    • If the count of input fields inside the request body is lesser than the values defined in the x-ibm-order property, all the input fields are rearranged.
    • If the input fields come from Query, Path, and Request body parameters, all the input fields are rearranged based on the availability of the parameter type in the skill specification.
    OpenAPI builder     		For the IBM Cloud offering, the OpenAPI builder doesn't work in the Toronto (ca-tor) and London (eu-gb) regions. If you try to access the OpenAPI builder in this region, an error page is displayed.
    Skills schemas Schemas that are defined by using anyOf, oneOf, and allOf are currently not supported.

    Building skills from apps

    Area affected Description
    Discovering skills through IBM Cloud Pak for Business Automation You might notice timeout issues while discovering apps by using IBM Cloud Pak for Business Automation as a service and On-premises. The issue can be caused if there are too many automations.
    Connecting to apps watsonx Orchestrate currently does not support insecure communication with servers that use self-signed certificates or any certificates not signed by a trusted certificate authority (CA).
    If you get the following error message, ask the administrator of the server to change the certificate to a certificate signed by a trusted CA:

    You couldn't connect. An error occurred while processing the request. Error: unable to get local issuer certificate.
    Member credential type option for skill-based actions A skill set to use personal credentials fails to run in AI assistants with the "You aren't connected" error even if the app is properly connected. Only Team credentials are allowed to be used with skill-based actions.

    Chat

    Following are the known issues and limitations that are applicable to the legacy chat and AI chat in the skill-based experience.

    Legacy chat

    The following table describes the known issues and limitations that are specific to the legacy chat:

    Area affected Description
    Getting content from files with skills If you use skills that get the contents from files like the Get file content skill from the Box app, or skills from workflows that have variables with Document type, you receive the file contents as plain text or binary depending on the file extension.

    Text files that have .txt extension return the file content as a human-readable content (plain text) as output. However, for binary files, which include other file extensions like .docx and .pdf, the skill returns the content encoded as Base64.

    As a workaround, you can use a Base64 decoder to decode the content to plain text.

    AI chat

    The following table describes the known issues and limitations that are specific to the AI chat:

    Area affected Description
    Starting skills with direct mention (@) to AI assistants When a skill is called by using the user's prompt with direct mention (@) to an AI assistant, the default AI assistant (watsonx) might not associate the inputs that are given for the skill in the users' prompt with the wanted skill.
    Instead, the default AI assistant (watsonx) might process the prompt by using the base LLM.
    Small talks might not work as expected The base LLM might not return the expected responses for small talks.
    Live agent support Live agent is not supported in the AI chat.
    Next best skill appears in the after chat even though its action is deleted from the AI assistant Skills added as next best skills appear as options in the chat even though their actions are deleted from the AI assistants that they belong to. As a workaround, you can remove the wanted skills from the next best skills directly in the skill enhancement.
    Data size limitation in the AI chat The AI chat has a size limitation of 1 MB for data that is returned from the forms of skills.
    The LLM might not detect digression for input fields like 'description' The LLM interprets inputs based on descriptions of the input fields of the skills. For fields that take generic data as input, the model might not always enforce strict input validation. Fields like 'description' usually expects that users enter freeform content, and in this case the model intends to accept various textual inputs without detecting digressions.
    If you must start a new conversation unrelated to the ongoing skill, start the conversation on a new chat.
    The second slot for the input field is automatically taking the input data that is provided for the first slot IBM Granite models identify the input data to fill the slots for input fields by using the description defined for these input fields. When an input field does not have a description, the model automatically tries to fill the slots with the data it has.
    To improve the behavior of the model, builders must provide a clear and descriptive description for the input fields in the OpenAPI specification of the skill.

    Enhancing skills

    Area affected Description
    Output (Table and Text responses) When you use the standard Table response in your skill and decide to switch to a consistent Text response, your change works normally. However, if you want to switch again from Text to Table response, first, you must remove the text inserted into Text field, and then return to the Table response successfully.
    Adding or enhancing skills that are based on projects When you enhance a project based skill or adding such a skill to a skill set, you see a separate tile for each workflow based operation, and a single tile that contains all decision based operations for the project. To find a skill quickly, search for the operation or skill name instead of the application name.
    Inputs of the skill being enhanced are not generated as expected in the slot logic flow For the skill being enhanced, you might not see its input parameters generated as outputs of the slot logic flow as you might expect. If the input parameters are in the request body, then you might see just one output parameter called "body" with a complex type that is generated with the contained parameters. Other parameter types, such as header and query parameters, are generated as an output parameter of the logic flow with the expected name and type.
    Language support for the Auto-generated phrases feature The Auto-generated phrases feature supports the English language only.
    Enhancing an existing skill The skill creator alone can improve a skill. Any other user must initiate a new skill if modifications are required.

    Managing app connections

    Area affected Description
    Deleting the credential type of an app If the Credential type for the app is set to Team credentials, and the connection is deleted from the skill catalog, the Connected by option displays the name of the user who originally connected the app to the team. As a workaround, edit and update the connection type from the Manage team page.
    Team skill set is missing for new tenant The team skill set is missing for newly provided tenant. However, after the admin adds a user to the tenant, the team skill set becomes visible.

    Monitoring usage

    Area affected Description
    Skill sets created from AI assistant builder Admins can't monitor usage data of skills that are used in team skill sets created from environments in AI assistant builder.

    Skill catalog

    Area affected Description
    The search on the skill catalog page does not return the expected result The search functionality on the Skill catalog page aims to find exact matches or synonyms of the search keyword. Therefore, the search in the skill catalog page currently fails to return the required result for certain apps. For example, when you search for the monday.com app, by using keywords such as monday or monday.com does not yield the expected result. Instead, users must use the keyword mondaydotcom to obtain the app as the search result.
    Skills with Japanese characters are not searchable The search in the Skill catalog page doesn't find results for skills named with Japanese characters.

    Skill flow

    Area affected Description
    The Cancel option in a skill flow The Cancel button in a skill flow does not cancel the current step in your chat conversation but cancels the entire chat conversation.
    Updating filters while running a skill flow Currently, it is not feasible to update the filters that were chosen in the preceding step of your skill flow. To refresh the filters, you must start the skill flow from the beginning.
    Next best action for a skill flow When a skill flow has another skill flow set as the next best action, the skill flow does not automatically pre-populate the values based on the input or output from the previous skill. You must manually add the values.
    Updating a skill flow Updating a skill flow might be partially successful in some circumstances. As a builder, you might not be aware that the team skill set update failed. As a user, you might notice an error while running a skill flow if any component skill that was introduced in the recent update of the flow is not present in your personal skill set. To resolve the issue, add the concerned skill set to your skill set and then run the skill.
    Editing exported skill flow files Edits in exported specification files of skill flows are not supported on watsonx Orchestrate.
    Using decision skills in a skill flow When two custom decision skills are added consecutively in a skill flow, the necessary data mapping is not available and you can’t complete the flow.

    Prebuilt apps and skills

    Area affected Description
    Error connecting to a skill If you notice an error such as "The skill can't be used right now... Invalid refresh token", you must disconnect and reconnect the app to resolve the issue. You may notice the issue when you initially connect and then return after a long time. To resolve the issue, you must re-authenticate by connecting the apps again.
    Output format of productivity skills The output format of productivity skills such as Create an issue or Create a pull request on the GitHub app returns an output as a decimal number, rather than an integer.
    Duplicate skills appearing on watsonx Orchestrate home page An app update might not recognize that the skills were already added in the previous version. Therefore, it can lead to duplicate skills appearing on the watsonx Orchestrate home page.
    Using IBM Document processing app
    1. Upon clicking the Connect app, you are prompted to enter the Bearer Token, which is not required. As a temporary workaround, you can provide any fictional data to connect. The connection information to IBM Document Processing is handled by watsonx Orchestrate internally.
    2. To protect the pods that implement the service and make sure of the availability, IBM Document processing skills have some limitations about their usage. The following list describes these limitations.

      Documents can't exceed any of the following size limits:

      • Size more than 10 MB per document.
      • More than 600 pages per document.
      • If concurrent requests exceed the maximum capacity, the requests might fail.
    Connecting to the watsonx.ai app Upon clicking the Connect app, you are prompted to enter the Bearer Token, which is not required. As a temporary workaround, you can provide some dummy data to connect. The connection information to watsonx.ai is handled by watsonx Orchestrate internally.
    Upload a file skill on the Box app A file can only be up to 16 MB when uploaded as an attachment.
    Delete a folder skill on the Box app While you use the skill Delete a folder, set the Recursive parameter to true. It ensures that all files and subfolders within the folder are also deleted. If you notice that the skill is still not deleting the files or subfolders within the folder, then remove the skill and re-add it before trying again.
    Create a feed comment skill on the Salesforce app You might notice issues when you select the Draft option in the Status field. Refrain by using the Draft option until the issue is fixed.
    Create asset skill skill on the Salesforce app An error message "The Create an asset skill can’t be used right now. Retry your request later" might appear when you use the skill. Make sure you mentioned an Account ID or a valid Contact ID with an Account ID mapped to it, or both.
    Create a contact or Create an account skills on the Salesforce app An error message "The provided data for the Account object isn't in the correct format for the application. Ensure that the data provided for the operation is valid. Ensure that the properties of the object comply with the definition in the OpenAPI document.... " might appear when you use the skill. As a workaround, remove and re-add the skill to your skill set. Verify that the connection details have also been removed and added again.
    Create an order skill on the Salesforce app
    • An error message "We can't complete the CREATE action in Salesforce because an account ID or contract ID is required for the Order object. Ensure that you completed the Account ID or Contract ID field for this action" might appear when you use the skill. Make sure you mentioned an Account ID or a valid Contract ID with an Account ID mapped to it, or both.
    • You might also notice issues when you select the Activated option in the Status field. Refrain by using the Activated option until the issue is fixed.
    Adding a connection to a Salesforce skill You might notice an incorrect page error when attempting to connect to a Salesforce skill. Reach out to your builder and request them to set up a minimum of one Salesforce skill. This ensures that the Salesforce application page appears in the skill catalog where connection information is available.
    Connecting to apps watsonx Orchestrate saves the credentials that are provided to connect an app without validation, and running the skill displays an error message if the credentials are incorrect. You can edit the connection and specify the correct credentials to use the skill.
    Granting permission to Microsoft Outlook application Administrators getting errors while granting access to Microsoft Outlook must enable the option Allow users to request access to this application to provide the users with self-service application access. For more information, see Enable self-service application assignment.
    Updating a prebuilt skill When a prebuilt skill is updated, it might be necessary to remove the outdated skill and add the updated skill to your skill set.
    Skills that run delete actions are returning no records Some skills from prebuilt apps return the "No records were found. Try again later." response for delete actions even if the skills complete successfully.
    Language support for prebuilt apps and skills Prebuilt apps and skills are available in the English language only.
    Connecting to the Asana app watsonx Orchestrate fails to connect to the Asana app due to connection issues from Asana side. The issue is observed in the trial accounts that reach the maximum limit of connections and can result in failure.
    Skills that run delete actions from HubSpot CRM app The skills from the HubSpot CRM app that run delete actions return success even if the item to be deleted doesn't exist in the application.
    Connecting to the Salesforce Account Engagement app     		The connection to the Salesforce Account Engagement app is not available.
    Availability of prebuilt apps and skills     		Starting from 29 August, you can use the prebuilt apps and skills that are available for you in the product. However, if the tenant that you are member was provisioned before 29 August, the apps and skills won't be available.
    Data mapping for the ID input parameter of the Delete a file skill from the Box app You might receive a response with the status code "500" and the "Internal Server Error" message when you run a workflow that contains the Delete a file skill from the Box app, and this skill has a data mapping that is configured for the ID input parameter.

    As a workaround, instead of using the ID parameter, use the where parameter in the data mapping to define the ID of the file that you want to delete.

    For the where parameter, use the following syntax: {"id":"<FILE_ID>"}, where <FILE_ID> is the ID of the file in the Box application.

    This known issue happens when you run an instance of a workflow in the Preview mode.
    Adding skills that update actions from the Salesforce app throws error You might notice an error when you try to add skills that update actions from the Salesforce app. You must reconfigure the Salesforce app to make it work.
    Configuring prebuilt skills for the AI agent Salesforce skills currently appear on the skills and apps page for the AI agent even though the skills are not configured from Skill studio. However, before adding a skill to the AI agent with a routing description, ensure that the skill is configured from Skill studio. For more information, see Configuring prebuilt app endpoints and skills schemas.

    Skill catalog

    Area affected Description
    The search on the skill catalog page does not return the expected result The search functionality on the Skill catalog page aims to find exact matches or synonyms of the search keyword. Therefore, the search in the skill catalog page currently fails to return the required result for certain apps. For example, when you search for the monday.com app, by using keywords such as monday or monday.com does not yield the expected result. Instead, users must use the keyword mondaydotcom to obtain the app as the search result.
    Skills with Japanese characters are not searchable The search in the Skill catalog page doesn't find results for skills named with Japanese characters.

    Skill flow

    Area affected Description
    The Cancel option in a skill flow The Cancel button in a skill flow does not cancel the current step in your chat conversation but cancels the entire chat conversation.
    Updating filters while running a skill flow Currently, it is not feasible to update the filters that were chosen in the preceding step of your skill flow. To refresh the filters, you must start the skill flow from the beginning.
    Next best action for a skill flow When a skill flow has another skill flow set as the next best action, the skill flow does not automatically pre-populate the values based on the input or output from the previous skill. You must manually add the values.
    Updating a skill flow Updating a skill flow might be partially successful in some circumstances. As a builder, you might not be aware that the team skill set update failed. As a user, you might notice an error while running a skill flow if any component skill that was introduced in the recent update of the flow is not present in your personal skill set. To resolve the issue, add the concerned skill set to your skill set and then run the skill.
    Editing exported skill flow files Edits in exported specification files of skill flows are not supported on watsonx Orchestrate.
    Using decision skills in a skill flow When two custom decision skills are added consecutively in a skill flow, the necessary data mapping is not available and you can’t complete the flow.

    Parent topic:

    → Release notes