aichatweb template: generated project throws "generic key type is 'String', but the key property 'Key' has type 'Guid'"

[prabh-62]

Description

A project scaffolded with dotnet new aichatweb --provider azureopenai (template Microsoft.Extensions.AI.Templates 10.9.0-preview.3.26411.16, default --vector-store local, which now maps to CommunityToolkit.VectorData.SqliteVec) throws on the first chat/search request:

System.InvalidOperationException: The collection's generic key type is 'String', but the key property 'Key' has type 'Guid'. The generic key type must match the key property type.

The generated code is internally inconsistent about the vector-store key type:

Generated file	Key type used
Services/IngestedChunk.cs	public required Guid Key { get; set; }
Program.cs	builder.Services.AddSqliteCollection<string, IngestedChunk>(...)
Services/SemanticSearch.cs	VectorStoreCollection<string, IngestedChunk>
Services/Ingestion/DataIngestor.cs	new VectorStoreWriter<string>(...), new IngestionPipeline<string>(...)

The record's [VectorStoreKey] property is Guid; everything that binds the collection uses string as TKey. SqliteVec validates that these match (SqliteModelBuilder.ValidateKeyProperty) and throws when building the model.

Reproduction Steps

Reproduce via the template

1. dotnet new install Microsoft.Extensions.AI.Templates::10.9.0-preview.3.26411.16
2. dotnet new aichatweb --provider azureopenai -o ReproGuid
3. Set AzureOpenAI:Endpoint + chat/embedding deployment names, then dotnet run
4. Send any chat message. The exception is thrown from SemanticSearch the first
   time the collection is resolved. (dotnet build succeeds — this is runtime only.)

Unhandled exception. System.InvalidOperationException: The collection's generic key type is 'String', but the key property 'Key' has type 'Guid'. The generic key type must match the key property type.
   at Microsoft.Extensions.VectorData.ProviderServices.CollectionModelBuilder.ValidateKeyProperty(KeyPropertyModel keyProperty)
   at CommunityToolkit.VectorData.SqliteVec.SqliteModelBuilder.ValidateKeyProperty(KeyPropertyModel keyProperty) in /home/runner/work/AI/AI/MEVD/src/SqliteVec/SqliteModelBuilder.cs:line 25
   at Microsoft.Extensions.VectorData.ProviderServices.CollectionModelBuilder.ValidateProperty(PropertyModel propertyModel, VectorStoreCollectionDefinition definition)
   at Microsoft.Extensions.VectorData.ProviderServices.CollectionModelBuilder.Validate(Type type, VectorStoreCollectionDefinition definition)
   at Microsoft.Extensions.VectorData.ProviderServices.CollectionModelBuilder.Build(Type recordType, Type keyType, VectorStoreCollectionDefinition definition, IEmbeddingGenerator defaultEmbeddingGenerator)
   at CommunityToolkit.VectorData.SqliteVec.SqliteCollection`2..ctor(...)
   at CommunityToolkit.VectorData.SqliteVec.SqliteVectorStore.GetCollection[TKey,TRecord](String name, VectorStoreCollectionDefinition definition) in /home/runner/work/AI/AI/MEVD/src/SqliteVec/SqliteVectorStore.cs:line 69
   at Program.<Main>$(String[] args)

Expected behavior

The scaffolded project builds and runs; document ingestion and semantic search work out of the box.

Actual behavior

InvalidOperationException the first time the IngestedChunk collection is built, because the [VectorStoreKey] property is Guid while the collection is bound with string as TKey.

Regression?

Appears to be fallout from the DataIngestion key-type migration from string to auto-generated Guid keys (#7410, #7396, #7557). Services/IngestedChunk.cs was updated to Guid Key, but Program.cs / SemanticSearch.cs / DataIngestor.cs and the pinned Microsoft.Extensions.DataIngestion version were not.

Known Workarounds

Make the key type consistent. The only variant that compiles against the pinned
package set is string — change IngestedChunk.Key back to string:

[VectorStoreKey(StorageName = "key")]
[JsonPropertyName("key")]
public required string Key { get; set; }

Going the other way (AddSqliteCollection<Guid, IngestedChunk>, VectorStoreCollection<Guid, IngestedChunk>, VectorStoreWriter<Guid>, IngestionPipeline<Guid>) does not compile with Microsoft.Extensions.DataIngestion 10.9.0-preview.1.26411.16, because SemanticSimilarityChunker is IngestionChunker<string> there:

error CS1503: Argument 2: cannot convert from
'Microsoft.Extensions.DataIngestion.Chunkers.SemanticSimilarityChunker' to
'Microsoft.Extensions.DataIngestion.IngestionChunker<System.Guid>'

Configuration

• Template: Microsoft.Extensions.AI.Templates 10.9.0-preview.3.26411.16
• .NET SDK: 10.0.400
• --provider azureopenai, --vector-store local (SqliteVec), no Aspire
• Generated package references:
  • Microsoft.Extensions.AI / Microsoft.Extensions.AI.OpenAI 10.9.0
  • Microsoft.Extensions.DataIngestion / .Markdig 10.9.0-preview.1.26411.16
  • CommunityToolkit.VectorData.SqliteVec 1.0.1-preview
  • Microsoft.Extensions.VectorData.Abstractions 10.7.0 (transitive)
• OS: macOS 12.7.6 (x64). Isolated repro is OS-independent.

Other information

dotnet build on the generated project succeeds; the failure is purely at runtime when the SqliteVec model is built (eagerly in GetCollection, or lazily via SemanticSearch in the template).