Facing issues related to downstream error while creating conversation in Workflows.

Yogesh Dhavale (LTIMindtree Limited) 20 Reputation points Microsoft External Staff
2026-05-23T06:03:03.7166667+00:00

Hello ,

I am facing this error while trying to interact with workflows in new Azure AI Foundry

downstream_error{"error":"Error creating item on conversation","conversation_id":"conv_83409fdf3d9df0ba00xUC7HtWWfoYWTJ80V15EtlAJ1J3cB1BY","item_id":"","details":""}

Before it was working fine but facing it from 21st May 2026 onwards

Foundry Agent Service
Foundry Agent Service

A fully managed platform in Microsoft Foundry for hosting, scaling, and securing AI agents built with any supported framework or model

Answer accepted by question author
Anshika Varshney 15,625 Reputation points Microsoft External Staff Moderator
2026-05-25T01:29:12.0933333+00:00

Hi @Yogesh Dhavale (LTIMindtree Limited)

The “downstream error” you are seeing while creating the resource or deployment usually indicates a backend or dependency issue rather than a problem directly in your request. In most cases, it happens when one of the required services or configurations is not fully ready or fails during provisioning.

You can try the following checks to narrow down the issue.

First, verify the region you are using. Some services and models depend on backend availability in specific regions. If the region has limited capacity or temporary issues, it can cause downstream errors. Try using a different supported region and check if the issue persists.

Next, confirm that all dependent resources are properly created and active. For example, if your deployment depends on Azure AI Foundry, Cognitive Services, or linked storage, ensure all of them are in a healthy state and not in provisioning or failed state.

Also check your subscription and quota limits. If the required quota for that resource type or model is not available, the backend may fail and return a downstream error. You can review usage and limits from the Azure portal.

Another important step is to review the activity logs in the Azure portal. Activity logs usually provide more detailed error information which can help identify if the failure is due to authorization, quota, or backend issues.

You can also try deleting the failed deployment and creating it again. Sometimes provisioning fails due to temporary backend glitches, and retrying resolves the issue.

Additionally, ensure that you have the necessary permissions on the resource group and subscription. Missing roles or restricted access can also result in failures during resource creation.

If you are using API or SDK, verify that the request payload is correct and includes all required parameters. Incorrect or incomplete configurations can also trigger downstream processing errors.

For reference, you can check the official troubleshooting guidance here https://learn.microsoft.com/azure/azure-resource-manager/troubleshooting/

If possible, please share the exact error message, region, and resource type you are trying to create. That will help to identify the root cause more precisely.

Thankyou!

Was this answer helpful?

2 people found this answer helpful.

3 additional answers

Sort by: Oldest
Most helpful Newest Oldest

  1. AI answer

    2026-05-23T06:03:30.12+00:00

    The error message indicates a failure when creating an item on an existing conversation in the new Azure AI Foundry / Azure OpenAI conversations API. From the provided context, there is no specific documented downstream_error condition or dated incident, but the following checks and remediation steps are supported by the documented APIs and troubleshooting guidance:

    1. Verify that the correct client and API are used
    • For the new agents/conversations experience, ensure the OpenAI client from the project is used, not the project client directly.
    • In Python, for example, instead of calling project_client.conversations.create(), obtain the OpenAI client first and then call conversations on it:
      • Use project.get_openai_client() and then openai.conversations.create().
    • If AttributeError: 'AIProjectClient' has no attribute 'conversations' or similar errors were previously seen and fixed, re‑confirm that the code still follows this pattern after any SDK updates.
    1. Confirm the correct endpoint and API version
    • For Azure OpenAI in Foundry Models v1:
      • endpoint must be a supported Azure OpenAI endpoint such as https://{your-resource-name}.openai.azure.com.
      • If api-version is used, ensure it is set to v1 or omitted (defaults to v1).
    • For Microsoft Foundry project endpoints:
      • Use the project endpoint in the form https://{ai-services-account-name}.services.ai.azure.com/api/projects/{project-name} or .../api/projects/_project for the default project.
    • Mismatched or incorrect endpoints can cause downstream errors when the service attempts to create conversation items.
    1. Validate conversation and item operations
    • Creating a conversation:
      • POST {endpoint}/openai/v1/conversations
      • Request body can include:
        • items: array of OpenAI.InputItem (optional in Foundry project API, required in some OpenAI endpoints).
        • metadata: up to 16 key–value pairs.
    • Creating conversation items:
      • POST {endpoint}/openai/v1/conversations/{conversation_id}/items
      • Required body: items (array of OpenAI.InputItem / OpenAI.Item depending on the API flavor).
      • Up to 20 items can be added at a time.
    • Ensure:
      • conversation_id exists and is valid (e.g., verify with GET {endpoint}/openai/v1/conversations/{conversation_id}).
      • The items array is present and correctly structured.
      • No more than 20 items are sent per request.
    1. Check for invalid or deleted conversations/items
    • If a conversation was deleted via:
      • DELETE {endpoint}/openai/v1/conversations/{conversation_id} then subsequent attempts to add items to that conversation can fail.
    • Similarly, if items are being manipulated (created, deleted) in quick succession, confirm that the conversation_id and any item_id used are current and not removed.
    1. Re‑test with a minimal request
    • Create a brand‑new conversation with a minimal payload (e.g., a single simple InputItem).
    • Then immediately call create conversation items on that new conversation.
    • If this minimal flow works, the issue is likely with the structure or content of the items being sent in the failing scenario.
    1. Consider SDK / platform changes
    • The context notes that the new agents developer experience and related SDKs have undergone breaking changes (e.g., create_agent() removed, new create_version() methods, updated packages like @azure/ai-projects@2.0.0).
    • If the issue started suddenly around a specific date without code changes, verify whether:
      • SDK versions were updated automatically (e.g., via package ranges).
      • The project or agents were migrated to the new experience.
    • After migration, old thread or conversation data is not migrated; new conversations must be started. Old data remains only via the previous API until deprecation. Ensure the code is not mixing old and new APIs.
    1. If using language SDKs, align versions and dependencies
    • For JavaScript: ensure @azure/ai-projects is 2.0.0 or later if using the new agents/conversations features.
    • For C#: ensure Azure.AI.Projects is installed (it brings in Azure.AI.Extensions.OpenAI and Azure.AI.Projects.Agents).
    • For Java: ensure com.azure:azure-ai-agents:2.0.0 is referenced.
    • Misaligned or outdated SDKs can cause runtime errors that surface as downstream failures when calling the REST API.

    If, after these checks, the downstream_error persists for all conversations (including newly created ones) and there were no code changes, this may indicate a service-side regression or incident. In that case, capture:

    • The apim-request-id response header from failing calls.
    • The exact endpoint and payload.
    • The time window of failures.

    Then open a support ticket with Azure support, providing those details so the service team can investigate the downstream error on the backend.

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

  2. kagiyama yutaka 5,570 Reputation points
    2026-05-24T05:45:21.8933333+00:00

    I think downstream_error means the service failed to create the conversation item, and Azure does not publish any user‑side cause. Create one new conversation and send a message, and when the same error appears, collect the apim‑request-id and contact Azure Support.

    Was this answer helpful?

    0 comments
  3. Sina Salam 31,456 Reputation points Volunteer Moderator
    2026-05-26T18:13:49.77+00:00

    Hello Yogesh Dhavale (LTIMindtree Limited),

    Welcome to the Microsoft Q&A and thank you for posting your questions here.

    I understand that you are facing issues related to downstream error while creating conversation in Workflows.

    Treat this as a service-side Workflow runtime regression in Microsoft Foundry’s conversation-item creation path unless you can prove otherwise with a minimal reproduction. Therefore,

    • Do not keep editing workflow nodes trying random payload/schema changes.
    • Do not assume the conversation_id is the root cause.
    • Do not spend more time on endpoint/client tuning unless your issue is only in custom SDK code and not in the Foundry workflow UI.

    On your own side for now:

    1. Create Azure support escalation via your portal - https://learn.microsoft.com/en-us/azure/azure-portal/supportability/how-to-create-azure-support-request or use Priority Customer Support (PCS) - https://learn.microsoft.com/en-us/azure/azure-portal/supportability/priority-community-support
    2. Workaround is to bypass Workflows and use direct agent/responses operations if that path still works. - https://learn.microsoft.com/en-us/azure/foundry/observability/how-to/trace-agent-setup, https://learn.microsoft.com/en-us/azure/foundry/reference/foundry-known-issues, https://learn.microsoft.com/en-us/azure/ai-services/cognitive-services-support-options That's if the business cannot wait for the workflow service fix, bypass the workflow layer and invoke a prompt agent / response flow directly. Microsoft documents that Foundry Agent Service supports agents, conversations, and responses, and that a conversation is optionally, you can generate responses directly, or maintain history through response chaining/client-managed context. - https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/runtime-components, https://learn.microsoft.com/en-us/azure/foundry/quickstarts/get-started-code

    I hope this is helpful! Do not hesitate to let me know if you have any other questions, steps or clarifications.

    Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.

    Was this answer helpful?

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.