Mem0Sharp 0.1.2

Mem0Sharp

Long-term memory for AI applications in .NET 10. Mem0Sharp is an independent C#/.NET implementation of the open-source Mem0 project, with one service API for saving, searching, updating, and deleting semantic memories while keeping embedding and storage providers replaceable.

Mem0Sharp is not affiliated with, sponsored by, or endorsed by Mem0 or mem0ai.

Attribution and trademarks

Mem0Sharp is an independent .NET implementation inspired by the open-source Mem0 project. The original Mem0 project is copyright 2023 Taranjeet Singh and is licensed under the Apache License 2.0. Copyright for the Mem0Sharp implementation and its modifications is held by Jihad Khawaja and contributors. See NOTICE and LICENSE for the complete attribution and license terms.

Mem0 and related names, logos, and marks belong to their respective owners. This project does not grant or claim any trademark rights and is not affiliated with, sponsored by, or endorsed by the Mem0 project or mem0ai.

Documentation

• Getting started - install, create a service, and use the core API.
• Providers and persistence - configure OpenAI-compatible embeddings, LLM extraction, and PostgreSQL with pgvector.
• API reference - understand models, filters, scopes, options, and extension points.

Features

• Semantic memory search with configurable result limits.
• CRUD operations plus filtered bulk deletion.
• User, session, and agent scopes with user, agent, and run filters.
• Metadata attached to each memory.
• Zero-dependency in-memory storage for tests and local development.
• Deterministic local embeddings for offline development.
• OpenAI-compatible chat completion and embedding support.
• PostgreSQL persistence with pgvector and optional HNSW indexing.

Quick start

The default service uses InMemoryStore, LocalEmbeddingGenerator, and BasicMemoryExtractor. It is a good starting point for development and tests; use a persistent store in production.

using Mem0Sharp;

var memory = new MemoryService();
await memory.AddAsync("I prefer dark mode and vim keybindings", userId: "alice");

var results = await memory.SearchAsync(
	"What editor settings does Alice prefer?",
	new MemoryFilter(UserId: "alice"),
	topK: 3);

foreach (var result in results)
{
    Console.WriteLine($"{result.Score:F3}: {result.Memory.Text}");
}

var allAliceMemories = await memory.GetAllAsync(new MemoryFilter(UserId: "alice"));
var memoryId = allAliceMemories[0].Id;
await memory.UpdateAsync(memoryId, "I prefer dark mode and Vim keybindings");
await memory.DeleteAsync(memoryId);

To add memories extracted from a conversation, pass Message values. The default extractor stores non-empty message content; the LLM-backed extractor is shown in the provider guide.

await memory.AddAsync(
[
    new Message("user", "I live in Berlin."),
    new Message("assistant", "Thanks, I will remember that.")
],
userId: "alice",
scope: MemoryScope.User);

OpenAI-compatible provider

Pass an HttpClient whose BaseAddress points at the provider root, such as https://api.openai.com/:

var provider = new OpenAiCompatibleClient(httpClient, Environment.GetEnvironmentVariable("OPENAI_API_KEY")!);
var memory = new MemoryService(
	embeddings: provider,
	extractor: new LlmMemoryExtractor(provider));

PostgreSQL with pgvector

Install PostgreSQL with the vector extension, then initialize a store using the same embedding dimension as the configured embedding model:

var store = new PostgresMemoryStore(new PostgresMemoryStoreOptions
{
	ConnectionString = Environment.GetEnvironmentVariable("MEM0_POSTGRES")!,
	EmbeddingDimensions = 1536,
	TableName = "mem0_memories"
});
await store.InitializeAsync();

var memory = new MemoryService(store, provider, new LlmMemoryExtractor(provider));

The store persists memory metadata and embeddings, applies user/agent/run/scope filters in SQL, uses cosine distance for vector search, and creates an HNSW index when the embedding dimension is supported by pgvector. CreateExtension = false can be used when the database user cannot create extensions.

MemoryService also provides SearchManyAsync for batch queries and DeleteAllAsync for filtered bulk deletion.

The provider boundary also works with compatible local servers. See Providers and persistence for model selection, embedding dimensions, and initialization details.

Install from NuGet.org

dotnet add package Mem0Sharp

Build and test

dotnet build .\src\Mem0Sharp\Mem0Sharp.csproj
dotnet test .\tests\Mem0Sharp.Tests\Mem0Sharp.Tests.csproj