Cannot create vector index in Microsoft Foundry – On Your Data Ingestion API retired

Question

Cannot create vector index in Microsoft Foundry – On Your Data Ingestion API retired

Nienke 0

I am unable to update an existing vector index in Microsoft Foundry.

I am using the built-in functionality in Microsoft Foundry to update an existing data source and its vector index. This workflow previously worked, but now fails with the following error:

404 NotFound

"The Azure OpenAI - On Your Data Ingestion APIs have been retired. This endpoint is no longer available. For migration guidance, see aka.ms/oyd-migration"

The provided migration URL redirects to Bing instead of providing migration instructions.

This error occurs directly within Microsoft's own Foundry interface, not through custom code.

I have been using this functionality for an existing solution and was not aware of any retirement announcement or required migration.

Could you please clarify:

Is this a known issue with the Microsoft Foundry interface?

How can I continue updating my existing vector index?

Is there a supported replacement workflow that preserves my existing configuration?

Was this retirement communicated to existing users, and if so, where?

I need to keep my existing solution operational and update its data without rebuilding everything.

Thank you for your assistance.

Balantrapu Sreevishnu 90 Reputation points Microsoft External Staff

2026-11-03T17:21:42.1866667+00:00
Hello @Nienke ,

Thank you for the detailed description. The 404 isn't a fault in your index or project. Azure OpenAI On Your Data retired on October 14, 2026. That includes its ingestion APIs and the related "add/update data" workflow in the Foundry portal, so this error is now expected behavior.

How to keep updating your existing vector index (no rebuild needed):

Your index is stored in your Azure AI Search service, not in On Your Data, so you can keep using it:

Open your Azure AI Search resource in the Azure portal and check Indexes, Data sources, Indexers and Skillsets for the index that On Your Data created.

If an indexer exists: upload your new or changed files to the same Blob container, then run the indexer (or keep its schedule). This refreshes the existing index.

If no indexer exists (for example, the index was built from local file uploads): create an indexer with an integrated vectorization pipeline on that index. You can use the Import and vectorize data wizard or the REST API/SDK, using the same embedding model so the vectors stay compatible.

Supported replacement for querying (keeps your existing index)

Foundry IQ knowledge base: wrap your existing index as a Search Index Knowledge Source, attach it to a knowledge base and call it directly. Add Foundry Agent Service if you also need tools or workflows.

Classic RAG: query Azure AI Search from your app and pass the results to a supported chat model.

Please note that migration isn't automatic, and these options don't work exactly like On Your Data. Review your field mappings, vectorizer, access control and citation handling before switching over. Requests that use data_sources or /extensions/chat/completions also need to be updated.

References

Retirement notice and migration paths: https://learn.microsoft.com/azure/foundry-classic/openai/concepts/use-your-data

Step-by-step migration sample (reuses your existing index): https://microsoft-foundry.github.io/forgebook/notebook/migrate-oyd-to-foundry-iq/

About your other questions:

Known issue? The retirement itself is by design. The aka.ms/oyd-migration link redirecting to Bing is a broken link, and we've noted it. Please use the links above instead.

Communication: The retirement is published on the Microsoft Learn page above. Retirement notices usually also go to subscription owners through Azure Service Health (Health advisories) and email, so you may want to check those and make sure your notification contacts are up to date.

If you share how your index was originally ingested (Blob, upload or URL), I can help confirm the exact refresh steps.

If this helps, please consider accepting the answer so it can help others with the same issue.

2 answers

Your answer

Answer 1

The classic Foundry On Your Data ingestion workflow can no longer update the index; maintain the existing Azure AI Search index directly, then migrate retrieval to Foundry IQ using Microsoft’s index-reuse sample.

What the 404 means

Azure OpenAI On Your Data retired on October 14, 2026, including its ingestion endpoints and classic portal workflows. The error is therefore expected retirement behavior rather than evidence that your index is missing or damaged. The classic documentation is retained only for historical and migration guidance.^1^

The Foundry interface you used is still invoking that retired backend. Retrying it, changing permissions, or recreating its connection will not restore the endpoint.

Keep the existing index updated

First determine which Azure AI Search assets remain:

In Azure portal, open the Azure AI Search resource associated with the solution.
Open Indexes and confirm the existing index is present.
Open Indexers and look for the indexer feeding that index.
Open Data sources and confirm the original Blob Storage or other source connection remains.

Then use the applicable path:

If the indexer still exists

Update the files or records in the original data source.
Go to Azure AI Search → Indexers → your indexer → Run.
Check Execution history and confirm the run completed successfully.
Query the index from Search explorer to confirm the new content is searchable.

Keep or add an indexer schedule if updates should be automatic. Indexers can carry changed source content through extraction, chunking, embedding, and indexing when integrated vectorization is configured.^2^

If the indexer no longer exists

One-time On Your Data ingestion could remove the indexer after completion, whereas scheduled ingestion retained it. In that case, configure an Azure AI Search ingestion pipeline consisting of:

The source data source
A skillset for chunking and embeddings
The existing compatible search index
A new indexer

The Azure portal’s Import data workflow can help create an integrated-vectorization pipeline. Ensure that its field mappings, embedding model, and vector dimensions match the existing index before running it.

For content-only changes, preserve the schema and reload or incrementally update documents. mergeOrUpload is the normal incremental action; an indexer can perform synchronization automatically when the source supports change tracking.^3^

Do not modify the existing vector field’s dimensions or vector configuration merely to refresh content. Adding fields or new vector configurations can sometimes be done in place, but changing or deleting existing fields often requires a full rebuild.^4^

Migrate without rebuilding the index

For an Azure AI Search-backed application, Microsoft’s replacement is Foundry IQ. The published migration sample explicitly reuses an existing compatible Search index:

Migrate Azure OpenAI On Your Data to Foundry IQ

The migration is not automatic. Use the sample to:

Connect Foundry IQ to the existing Azure AI Search index.
Map the current content, title, URL, vector, filter, and security fields.
Replace the On Your Data retrieval call with Foundry IQ retrieval.
Update answer generation and citation handling.
Preserve conversation history in the application.
Test authorization, unsupported questions, latency, cost, and citation resolution before cutover.

The existing index can be retained, but connector behavior, access-control lists, embeddings, retrieval parameters, and application orchestration do not automatically transfer.^5^

The sample’s automatic retrieval and answer-synthesis path uses preview functionality. For production workloads requiring generally available components, the documented alternatives are Azure AI Search knowledge-base retrieval with application-managed answer generation or classic RAG; these are not feature-equivalent drop-in replacements.^1^

Retirement communication

The retirement is now documented prominently in the classic On Your Data concept, API-reference, and troubleshooting pages. There is no separate public “known issue” notice identified for this specific Foundry UI failure.

Microsoft documents Azure Resource Health and subscription-owner email notifications for model retirements, but that notification policy does not establish whether every customer received a separate notification for the On Your Data feature retirement.^6^

If the classic Foundry UI still presents the retired update action without warning, open an Azure support request and include the project, region, UTC timestamp, correlation/request ID, and screenshot. This can address the stale portal experience, but the supported operational path remains direct Azure AI Search maintenance plus migration to Foundry IQ.

References

Answer 2

Hi @Nienke

This is not a transient vector-index failure. Microsoft retired Azure OpenAI On Your Data on October 14, 2026, including its ingestion APIs and the former Foundry portal workflow. A request to that endpoint now returning 404 NotFound is therefore expected.

The incorrect aka.ms/oyd-migration redirect appears to be a documentation-link issue. You can report it through Microsoft Learn feedback.

You don’t necessarily need to rebuild the existing vector index. A compatible Azure AI Search index created by file upload, Blob Storage, or web ingestion can be reused. However, migration isn’t automatic: field mappings, connectors, access-control behavior, embeddings, citations, and application calls must be reviewed or reconfigured.

For an existing Azure AI Search-backed solution:

Check the Azure AI Search resource for the existing index, data source, indexer, and skillset.
If a scheduled indexer exists, update the underlying Blob data and run or retain that indexer to refresh the current index.
If no indexer exists, create and manage an Azure AI Search ingestion pipeline separately. The former On Your Data ingestion service can no longer update it.
Connect the existing index to either:

A Foundry IQ knowledge base for managed retrieval and grounded answers; or
The Azure AI Search tool in Foundry Agent Service if the solution requires an agent, tools, or workflow orchestration.

The Agent Service Azure AI Search tool can connect to an existing vector index.

For production workloads, note that Microsoft’s migration sample uses preview capabilities for automatic retrieval and answer synthesis. Microsoft recommends either the generally available Foundry IQ retrieval API with application-managed answer generation or a classic RAG implementation when preview features aren’t acceptable.

Regarding notification, the retirement is now documented on the On Your Data concept, configuration, best-practices, and API-reference pages.

References:

Help make this community better for everyone: If this answer helped or resolved your issue, please accept it or upvote it. If not, share more details in a comment so we can continue the discussion and find the right solution. Thank you.