How to handle async manager approval (HITL) while still allowing user interaction in LangGraph?

I’m building an agent using LangChain + LangGraph for a entertainment industry.
When a user requests a large offer amount, the issuance must pause until a manager approves it. The manager approval is handled outside the agent, and when approved, an event is pushed into event hub.

I’m currently using an Interrupt in LangGraph to pause the workflow and wait for the external approval event.

However, during the approval wait time, the user should still be able to chat with the agent and ask other unrelated questions.

My concern:
If I use Interrupt on the main thread, it seems like the entire agent stops and the user cannot continue the conversation.

How can this flow be implemented?..

1 Like

Is it possible to run a multi-agent system where each agent graph executes in a separate thread (or async task), such that one agent’s graph can be interrupted, paused, or cancelled without affecting the others? Is there a supported way to interrupt a single agent graph execution mid-run?

Hi @bharathiselvan

is that sth helpful Stopping endpoint for deep agents - #2 by pawel-twardziak?

Have you managed to handle your original issue in this post?

Thanks @pawel-twardziak , The original issue was handled without interrupt.. I will check your post on stopping endpoint for deep agents and let you know.

1 Like