Unichrome 🚀

Unichrome is an embeddable vector storage for Unity.

Features 🌟

• Embeddable: Seamlessly integrates with your Unity project 🛠️.
• In-Memory Storage: Offers swift in-memory data handling ⚡.
• Local Persistence: Boasts local data saving with the speedy MemoryPack 📦.
• Embeddings:
  • Sentis Integration: Perfectly pairs with Unity Sentis for embedding creation 🤝.
  • Local: HuggingFace Sharp-Transformers: Utilizes HuggingFace Sharp-Transformers for local embeddings (via SentenceTransformer) 🧠.
  • (coming soon) API: OpenAI: Connects with OpenAI API for embedding generation (needs API key & Azure settings) 🔑.

Dependencies 📚

Props to:

• HNSW by Microsoft & Curiosity AI - top-notch work! 👏
• Sentis ML runtime by Unity Technologies for local model inference 🧪
• MemoryPack for our primary storage backend 🗃️
• HuggingFace Sharp-Transformers for their SentenceTransformers embeddings 💬

Installation Guide 🛠️

Unity's package.json doesn't play nice with git dependencies. You'll need to manually tweak your manifest file.

Manifest Installation

{
  "dependencies": {
    "com.magicforge.unichrome": "https://github.com/pdeschain/Unichrome.git?path=/Packages/Unichrome",
    "com.huggingface.sharp-transformers": "https://github.com/huggingface/sharp-transformers.git"
  }
}

MemoryPack will need to be added to your project too - the preferred way is to use a .unitypackage provided on the github of the project here. This will save you the need to manually manage the System.Runtime.CompilerServices.Unsafe.dll file in your project.

How to Use

Unichrome Hello World

using System.Collections.Generic;
using System.Threading.Tasks;
using Unichrome.Embeddings.HuggingFace;

namespace Unichrome.Sample
{
    public class UnichromeHelloWorld
    {
        public async ValueTask Example()
        {
            //set up SentenceTransformers embeddings
            var embeddings = new HuggingFaceSentenceTransformerEmbeddings();

            //set up UnichromeDB in-memory, for easy prototyping. Can add persistence easily by specifying a path and calling Persist()!
            var db = new UnichromeDB(embeddings);
            
            string collectionName = "HelloWorld";
            
            // Create collection. GetCollection, GetOrCreateCollection, DeleteCollection also available!
            var collection = db.GetOrCreateCollection(collectionName);
            
            // Supports adding, updating, getting and deleting documents.
            var id = await collection.AddDocumentAsync(
                text: "This is document1",
                metadata:new Dictionary<string, string>
                {
                    {"source", "notion"},
                });
            
            //Metadata can be used to filter search results.
            // 
            
            // Query/search 2 most similar results. 
            var results = await collection.SearchAsync("This is a query document", 2);
            
            // Query/search 2 most similar results with a filter on the metadata.
            var metadataSearchResults = await collection.SearchAsync("This is a query document", 2,
                new []{("source", "==", "notion")});
            
            //...when you're done, (optionally) persist the db and dispose the embeddings
            
            db.Persist();
            embeddings?.Dispose();
        }
    }
}