Discussion: Durable Conversation Memory / Thread State for Microsoft.Extensions.AI

[achawla83]

Hi team,

I have been exploring Microsoft.Extensions.AI and really like the lightweight/provider-agnostic design around IChatClient and the middleware pipeline.

While working with agentic AI workflows, I noticed there currently does not appear to be a built-in abstraction for durable conversation memory/thread state management similar to what frameworks like LangGraph provide.

Examples of the scenarios I’m referring to:

• automatic chat history persistence
• thread-based conversation restore
• checkpointing/resume
• workflow state persistence
• token-aware conversation trimming/summarization
• middleware-driven memory loading/saving

Today this seems to be implemented manually by consumers around IChatClient.

I wanted to ask:

1. Is the current direction intentionally keeping Microsoft.Extensions.AI stateless?
2. Is there any planned work around conversation memory or durable agent state abstractions?
3. Would the team prefer these capabilities to exist as external extension packages/middleware rather than inside the core library?
4. Would middleware-based conversation memory extensions align with the intended architecture of the library?

I’m interested in experimenting with a lightweight middleware/provider model for durable chat history and thread memory built around IChatClient, but wanted to first understand the intended direction before exploring this further.

Thanks!

[svankleef]

I came here exactly for this. Hope this gets implemented.