Implementation:Microsoft Semantic kernel InMemoryVectorStore

Overview

This page documents the InMemoryVectorStore class and the GetCollection method for creating typed vector store collections in Microsoft Semantic Kernel. The in-memory vector store is the simplest backend, ideal for development, testing, and prototyping, but the collection access pattern is identical across all supported backends.

Source Reference

File: dotnet/src/VectorData/InMemory/InMemoryVectorStore.cs (lines 19-100)
Type: API Doc

API Reference

`InMemoryVectorStore` Constructor

var vectorStore = new InMemoryVectorStore();

Aspect	Detail
Namespace	`Microsoft.SemanticKernel.Data`
Returns	`InMemoryVectorStore` instance
Parameters	None required
Thread safety	Thread-safe for concurrent collection access

The InMemoryVectorStore stores all data in process memory using concurrent dictionaries. Data is not persisted across application restarts. For production scenarios, use a persistent backend such as AzureAISearchVectorStore or QdrantVectorStore.

`GetCollection<TKey, TRecord>`

VectorStoreCollection<string, Glossary> collection = vectorStore.GetCollection<string, Glossary>("skglossary");

Aspect	Detail
Returns	`VectorStoreCollection<TKey, TRecord>`
Type parameters	`TKey` — the key type (e.g., `string`); `TRecord` — the record class with vector store attributes
Parameters	`collectionName` (`string`) — the name of the collection
Behavior	Returns a client-side collection reference; does not create backend resources

`EnsureCollectionExistsAsync`

await collection.EnsureCollectionExistsAsync();

Aspect	Detail
Returns	`Task`
Behavior	Creates the collection in the backend if it does not already exist; no-op if it exists
Idempotent	Yes — safe to call multiple times

Complete Setup Example

The following example demonstrates the full setup flow from store creation to a collection ready for data operations:

using Microsoft.SemanticKernel.Data;

// Step 1: Create the vector store
var vectorStore = new InMemoryVectorStore();

// Step 2: Get a typed collection
var collection = vectorStore.GetCollection<string, Glossary>("skglossary");

// Step 3: Ensure the collection exists in the backend
await collection.EnsureCollectionExistsAsync();

// The collection is now ready for UpsertAsync, SearchAsync, etc.

Backend Swappability

The same collection access pattern works across all backends. Only the store instantiation changes:

// In-memory (development)
var vectorStore = new InMemoryVectorStore();

// Azure AI Search (production)
var vectorStore = new AzureAISearchVectorStore(
    new Uri("https://my-search.search.windows.net"),
    new AzureKeyCredential("my-api-key"));

// Qdrant (self-hosted)
var vectorStore = new QdrantVectorStore(
    new QdrantClient("localhost", 6334));

After store creation, the collection access code is identical:

var collection = vectorStore.GetCollection<string, Glossary>("skglossary");
await collection.EnsureCollectionExistsAsync();

Collection Naming

The collectionName parameter maps to different backend concepts:

Backend	Collection Name Maps To
InMemory	Dictionary key
Azure AI Search	Index name
Qdrant	Collection name
Pinecone	Index name
Weaviate	Class name

Collection names should use lowercase alphanumeric characters and avoid special characters to ensure compatibility across all backends.

Type Parameter Constraints

The TRecord type parameter must be a class that:

Has exactly one property with [VectorStoreKey]
Has at least one property with [VectorStoreVector] for search scenarios
Uses supported property types for data fields

The TKey type parameter must match the type of the property marked with [VectorStoreKey] in the TRecord class.

Error Scenarios

Scenario	Behavior
`TKey` does not match key property type	Runtime exception
`TRecord` missing `[VectorStoreKey]`	Runtime exception on collection creation
Collection name contains unsupported characters	Backend-specific error during `EnsureCollectionExistsAsync`

Relationship to Principle

This implementation page corresponds to the Vector Store Collection Setup principle, which explains the factory pattern and provider abstraction philosophy.

Principle:Microsoft_Semantic_kernel_Vector_Store_Collection_Setup

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment