Add a Chroma provider (CommunityToolkit.VectorData.Chroma) #57

Summary

Add a Chroma provider for Microsoft.Extensions.VectorData as CommunityToolkit.VectorData.Chroma, in MEVD/src/Chroma with its unit and conformance tests, like the other providers of this repository.

Chroma is listed as planned for .NET in the vector store tables of Semantic Kernel and Agent Framework.

Background

In microsoft/semantic-kernel#12510, @roji wrote that a Chroma provider first needed an up-to-date low-level .NET client. ChromaDotNet.Client (repository) continues ChromaDB.Client and is tested against Chroma 0.4.10 to 1.5.9 and Chroma Cloud.

The provider

A working provider is at ChromaDotNet/ChromaDB.VectorData, published as ChromaDotNet.VectorData. It started as a copy of the Qdrant provider of this repository and keeps its structure.

• The Microsoft.Extensions.VectorData conformance tests pass against Chroma 1.5.0, 1.5.9 and the latest release, and on Chroma Cloud, hybrid search included. The test bases for multiple vectors and records without vectors are excluded: Chroma stores exactly one embedding per record.
• Limits that come from Chroma: metadata has no null values, so filtering on null is not supported; GetAsync with a filter does not support ordering; comparisons work on numbers only.

Change

A pull request that adds MEVD/src/Chroma, MEVD/test/Chroma.UnitTests and MEVD/test/Chroma.ConformanceTests with the namespaces, package metadata and test layout of the other providers. The conformance tests start Chroma with Testcontainers, so the mevd.yml matrix runs them with the other providers.

Once CommunityToolkit.VectorData.Chroma is published, ChromaDotNet.VectorData is deprecated in its favor on NuGet. We keep maintaining the provider here.