Progress.Memory.Extensions.AI 0.2.1

Progress.Memory.Extensions.AI

Microsoft.Extensions.AI integration for Progress Memory.

Adds automatic memory injection and LLM-callable memory tools to any IChatClient-based pipeline — OpenAI, Azure OpenAI, Ollama, or any other M.E.AI-compatible provider.

Not using Microsoft.Extensions.AI?
For Semantic Kernel agents, use Progress.Memory.SemanticKernel instead.

Requirements

• .NET 8, 9, or 10
• Progress.Memory (auto-included as a dependency)
• Microsoft.Extensions.AI ≥ 10.6.0

Installation

dotnet add package Progress.Memory.Extensions.AI

Integration modes

Choose one or combine both:

Mode 1 — Middleware (auto-inject)

Wrap any IChatClient with the UseProgressMemory middleware. On every call:

1. The last user message is used to search for relevant past memories.
2. Matching facts are prepended as a system message before the request reaches the LLM.
3. If autoRemember: true, the completed exchange is saved to memory (awaited inline) after the response.

Minimal example

// Register the memory client
builder.Services.AddProgressMemory(new ProgressMemoryConfig(
    BackendBaseUrl: "https://memory.example.com",
    SdkApiKey: "pmem_p_..."));

// Build an IChatClient with memory middleware
builder.Services.AddChatClient(pipeline => pipeline
    .UseProgressMemory(sp, userId: "user-123", agentId: "my-agent", autoRemember: true)
    .Use(new OpenAIChatClient(openAIClient, "gpt-4o")));

Dynamic identity (multi-tenant apps)

Use delegate overloads so identity is resolved at request time rather than at startup:

builder.Services.AddChatClient(pipeline => pipeline
    .UseProgressMemory(
        sp,
        getUserId:  () => httpContextAccessor.HttpContext!.User.FindFirst("sub")!.Value,
        getAgentId: () => "my-agent",
        autoRemember: true)
    .Use(new OpenAIChatClient(openAIClient, "gpt-4o")));

Parameters

Mode 2 — Tools (LLM-controlled)

Attach remember, recall, and forget as AIFunction tools so the LLM can invoke them autonomously when it decides memory is relevant.

var options = new ChatOptions()
    .WithProgressMemoryTools(memoryClient, userId: "user-123", agentId: "my-agent");

var response = await chatClient.CompleteAsync(messages, options);

The LLM schema describes each tool so the model knows when and how to use them.

Tool descriptions

Combining both modes

Use the middleware for seamless silent recall and add tools for LLM-directed write/delete:

builder.Services.AddChatClient(pipeline => pipeline
    .UseProgressMemory(sp, userId: "user-123", agentId: "my-agent")  // silent recall
    .Use(new OpenAIChatClient(openAIClient, "gpt-4o")));

// At call time, add tools for explicit remember/forget
var options = new ChatOptions()
    .WithProgressMemoryTools(memoryClient, userId: "user-123", agentId: "my-agent");
var response = await chatClient.CompleteAsync(messages, options);

Conversation threading

All turns within the same IConversationContext instance are linked into one conversation resource. Pass a shared ConversationContext instance across turns:

var context = new ConversationContext();

builder.Services.AddChatClient(pipeline => pipeline
    .UseProgressMemory(sp, userId: "user-123", agentId: "my-agent",
                       conversationContext: context)
    .Use(...));

See also

• Progress.Memory — core client and DI registration
• Progress.Memory.SemanticKernel — Semantic Kernel integration