Reuse of existing AI search index (rag real time audio) - issue with vector / fields

[zxrzxr69-creator]

Hi,

Using actually with no issue with all the great features activated from "azure-search-openai-demo,"
Want to reuse ressources with this repo aisearch-openai-rag-audio, and get this error : "(InvalidVectorQuery) Field 'embedding' does not have a vectorizer defined in it's vector profile."
Check on Azure portal :

so i have add vectorizer found in the list :

but still get an error "azure.core.exceptions.HttpResponseError: () Could not complete vectorization action. The vectorization endpoint returned status code '401' (Unauthorized).": does this due to the need of using Search Service ManagedIdentity ?

If i set the parameter : AZURE_SEARCH_USE_VECTOR_QUERY="false" as mentionned in aisearch-openai-rag-audio repo
azure.core.exceptions.HttpResponseError: () The field 'id' in the search field list is not searchable.
Parameter name: searchFields
Code:
Message: The field 'id' in the search field list is not searchable.
Parameter name: searchFields

any solution ?

[pamelafox]

Hm, we should be adding a vectorizer with the latest code. searchmanager.py is the file that sets up the vectorizer for both existing and new indexes. Here's the code that updates for existing indexes:

            if embedding_field and not any(
                field.name == self.field_name_embedding for field in existing_index.fields
            ):
                logger.info("Adding %s field for text embeddings", self.field_name_embedding)
                existing_index.fields.append(embedding_field)
                if existing_index.vector_search is None:
                    raise ValueError("Vector search is not enabled for the existing index")
                if text_vectorizer is not None:
                    if existing_index.vector_search.vectorizers is None:
                        existing_index.vector_search.vectorizers = []
                    existing_index.vector_search.vectorizers.append(text_vectorizer)
                if (
                    text_vector_search_profile is None
                    or text_vector_algorithm is None
                    or text_vector_compression is None
                ):
                    raise ValueError("Text vector search profile, algorithm and compression must be set")
                if existing_index.vector_search.profiles is None:
                    existing_index.vector_search.profiles = []
                existing_index.vector_search.profiles.append(text_vector_search_profile)
                if existing_index.vector_search.algorithms is None:
                    existing_index.vector_search.algorithms = []
                existing_index.vector_search.algorithms.append(text_vector_algorithm)
                if existing_index.vector_search.compressions is None:
                    existing_index.vector_search.compressions = []
                existing_index.vector_search.compressions.append(text_vector_compression)
                await search_index_client.create_or_update_index(existing_index)

Can you run prepdocs again and see if it's hitting up the vectorizer path? If not, I recommend changing the conditions for the "if" to hit that up. Do you have the latest version of this repo?

[onestardao]

Thanks for sharing the detailed context. From what you described, this looks closest to Problem #5: Semantic ≠ Embedding — the vector field setup and the actual semantic intent are mismatched (e.g. embedding config exists, but searchFields are not aligned or not marked as searchable).

Would you like me to share a quick checklist of common failure points for this category (vectorizer registration, searchFields consistency, auth scope mismatches)? It might help you systematically debug similar cases while waiting for upstream fixes.

[zxrzxr69-creator]

Hi,

I c;lone the latest repo here, launch prepdocs.py (no changes on the index, already md5 files) and force in .env : AZURE_SEARCH_USE_VECTOR_QUERY="false" in the env backend (don't know why it was disappearing each time after start.ps1) => now it works (with some little customization) !

thanks @pamelafox for your suggestions

[zxrzxr69-creator]

Clearly, if I have a vectorizer, it should work also if I don't specify this variable AZURE_SEARCH_USE_VECTOR_QUERY="false", but I get a 401 unauthorized (even if I had both the IAM search resources side and the OpenAI side)... isn't it ?

[onestardao]

Great to hear the latest repo works now!
About the 401 Unauthorized when AZURE_SEARCH_USE_VECTOR_OVERRIDE isn’t set:

• AZURE_SEARCH_USE_VECTOR_OVERRIDE only toggles whether embeddings are done client-side vs by the Search vectorizer. It’s not an auth switch. A 401 almost always means the Search auth config doesn’t match what the app expects.

Quick checks

1. Make sure AZURE_SEARCH_ENDPOINT points to the service (e.g., https://<service-name>.search.windows.net), not an index URI.

2. Set the key the app expects: most samples require an Admin key for index/write ops. Using a Query key will 401 on create/update.

   • Try setting AZURE_SEARCH_API_KEY (or AZURE_SEARCH_ADMIN_KEY) to the Admin key from the Search service keys blade.

3. If you’re using Entra ID/Managed Identity instead of an API key, confirm the demo supports it and that the principal has Search Service Contributor or Search Index Data Contributor. Also ensure the app is running in “AAD mode” (env vars like AZURE_TENANT_ID, AZURE_CLIENT_ID, etc., as required by the sample).

4. Confirm the 401 is coming from Search (the URL in the error). If it’s from Azure OpenAI, double-check AZURE_OPENAI_ENDPOINT and AZURE_OPENAI_API_KEY/role assignments.

5. If possible, paste the exact failing request path (e.g., /indexes('{index}')/docs/search?api-version=...) and the header it’s using (api-key vs bearer token). That will pinpoint whether it’s admin vs query key or AAD token.

TL;DR: the override flag isn’t the cause; it’s most likely an admin vs query key (or AAD role) mismatch on the Search side.