All videos
0:00 / 0:00
ai

HTTP Channels for Managed Deep Agents

LangChain3 October 2026Watch on YouTube

Part of series

Ep. 13 · LangChain Deep Agents

Stap-voor-stap uitleg en tutorials over het bouwen, configureren en deployen van Deep Agents met het LangChain-framework.

View the series

What you'll learn

  • You learn when you actually need a custom HTTP channel instead of a custom UI on the LangGraph SDK.
  • You see how to set up an agent that receives inbound requests from services you don't control, such as HubSpot, Salesforce, or a texting provider.
  • You learn to write verify and parse functions to check and process incoming messages.
  • You see how to manage threads with a stable thread ID and hash phone numbers into a durable user ID.
  • You learn to send replies with a post hook and check the result in LangSmith and Context Hub.

Frequently asked questions

What does Nathan Drezner build in the demo?
He builds Reflex, an expense-reporting agent that receives receipt photos over iMessage, extracts the data, and stores it in durable memory backed by Context Hub.
Which parts are covered when building the HTTP channel?
Writing verify and parse functions, managing thread IDs per user, sending replies with a post hook, and checking the result in LangSmith.
How is a durable user ID created?
Phone numbers are hashed into a durable user ID so threads stay stable per user.
What role does Photon play in the demo?
Photon is introduced as the channel provider through which the agent receives receipts over iMessage.

Topics

langchainagentshttp-channelslangsmithdeveloper-tutorial

In this video

LangChainLangChainSalesforce

Read next

Description from the channel

Nathan Drezner, a product manager at LangChain, shows how to build a custom HTTP channel for a Managed Deep Agent so it can receive inbound requests from any service you don't control, like HubSpot, Salesforce, or a texting provider. He builds Reflex, an expense-reporting agent that receives receipt photos over iMessage through Photon, extracts the data, and stores it in durable memory backed by Context Hub. Along the way he covers writing verify and parse functions, managing thread IDs per user, sending replies with a post hook, checking the result in LangSmith, and when to reach for HTTP channels versus a custom UI on the LangGraph SDK. Chapters: 0:00 Custom HTTP channels for managed deep agents, explained 0:14 When you actually need a custom HTTP channel 0:24 The demo: a receipt-triage agent over text 1:04 Scaffolding the agent and its instructions 1:26 Storing the ledger in durable memory, then testing it 2:04 Adding the memory definition 2:20 Introducing Photon as the channel provider 3:10 Building the client, verifier, and parser 4:19 Managing threads with a stable thread ID 4:38 Hashing phone numbers into a durable user ID 5:27 Sending replies with a post hook 6:37 Deploying and testing the agent live 7:03 Checking traces in LangSmith and memory in Context Hub 7:33 Wrap-up: authorization, combining channels, and choosing a UI Resources: → HTTP channels docs: https://docs.langchain.com/langsmith/javascript/managed-deep-agents-channels-http → Managed Deep Agents overview: https://docs.langchain.com/langsmith/managed-deep-agents-overview → LangSmith: https://www.langchain.com/langsmith → Join the Managed Deep Agents waitlist: https://www.langchain.com/langsmith-managed-deep-agents-waitlist