[API Proposal]: VectorData.Abstractions

[jeffhandley]

Background and motivation

The VectorData.Abstractions APIs have been under development as part of Semantic Kernel. Alongside the Microsoft.Extensions.AI APIs, we aim to promote these Vector Data APIs as fundamental building blocks of the .NET AI strategy.

There are many APIs to review as part of the abstractions, and we also have concrete Connectors to review once we get through the abstractions.

API Proposal

• Abstractions - API View
• Concrete Implementations - APIView

API Usage

We will discuss API usage during the review meeting and show examples. The VectorDataIntegrationTests (microsoft/semantic-kernel) can also be referenced.

MEVDSample.cs

Alternative Designs

No response

Risks

No response

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-meta
See info in area-owners.md if you want to be subscribed.

[adamsitnik]

One last thing I want to get confirmed before we move on to the next part: the names of the option bags of the GetAsync methods.

We have three GetAsync methods:

• Task<TRecord?> GetAsync(TKey key, GetRecordOptions? options = default, CancellationToken cancellationToken = default);
• IAsyncEnumerable<TRecord> GetAsync(IEnumerable<TKey> keys, GetRecordOptions? options = default, CancellationToken cancellationToken = default);
• IAsyncEnumerable<TRecord> GetAsync(Expression<Func<TRecord, bool>> filter, int top, GetFilteredRecordOptions<TRecord>? options = null, CancellationToken cancellationToken = default)

1. The first one returns a single record, the second return multiple. But they both accept GetRecordOptions (plural) option bag. Is that OK?
2. The third one accepts a filter and is called GetAsync, but it accepts GetFilteredRecordOptions option bag. Is that acceptable (we had long debates around the naming of this particular option bag)?

[jeffhandley]

Closing this out as it was a placeholder issue for work tracked in another repository. The API Review meetings were conducted and feedback was captured.