BlazorMemory 0.1.0

BlazorMemory

AI memory layer for .NET — LLM-powered fact extraction, vector search, and persistent memory for Blazor and ASP.NET Core apps.

BlazorMemory gives your AI assistant a persistent, searchable memory. It automatically extracts facts from conversations, stores them as vector embeddings, and injects relevant memories into future prompts — so your assistant actually remembers the user.

How it works

User message → Extract facts → Embed → Store
                                         ↓
Next message → Embed query → Vector search → Inject memories → LLM

Quickstart

dotnet add package BlazorMemory
dotnet add package BlazorMemory.Storage.IndexedDb      # Blazor WASM (browser)
dotnet add package BlazorMemory.Embeddings.OpenAi
dotnet add package BlazorMemory.Extractor.OpenAi

Blazor WASM (Program.cs)

builder.Services
    .AddBlazorMemory()
    .UseIndexedDbStorage()
    .UseOpenAiEmbeddings(apiKey)
    .UseOpenAiExtractor(apiKey);

Server-side / EF Core (Program.cs)

dotnet add package BlazorMemory.Storage.EfCore

builder.Services
    .AddBlazorMemory()
    .UseEfCoreStorage<YourDbContext>()
    .UseOpenAiEmbeddings(apiKey)
    .UseOpenAiExtractor(apiKey);

Use in a component or service

public class ChatService(IMemoryService memory)
{
    public async Task<string> ChatAsync(string userMessage, string userId)
    {
        // 1. Retrieve relevant memories
        var memories = await memory.QueryAsync(userMessage, userId,
            new QueryOptions { Limit = 5, Threshold = 0.65f });

        // 2. Build system prompt with memories
        var context = string.Join("\n", memories.Select(m => $"- {m.Content}"));
        var systemPrompt = $"You are a helpful assistant.\n\nWhat you know:\n{context}";

        // 3. Call your LLM (OpenAI, Anthropic, etc.)
        var reply = await CallLlm(systemPrompt, userMessage);

        // 4. Extract and store new memories from the exchange
        await memory.ExtractAsync($"User: {userMessage}\nAssistant: {reply}", userId);

        return reply;
    }
}

Packages

Features

• Automatic fact extraction — LLM pulls structured facts from raw conversation
• Intelligent consolidation — deduplicates and updates existing memories instead of storing redundant facts
• Vector similarity search — finds semantically relevant memories for any query
• Staleness scoring — surfaces fresh memories over stale ones
• Zero backend option — IndexedDB adapter keeps everything in the user's browser
• Provider agnostic — swap OpenAI for Anthropic or any future provider

License

MIT