LangChain updates Managed Deep Agents with user memory

LangChain has officially launched Managed Deep Agents v0.8, an upgrade designed to simplify how engineering teams deploy and run production-grade AI agents. The release addresses four primary production challenges: agent memory, authentication, communication channels, and tool management. By packaging the Deep Agents harness with managed infrastructure, the platform allows developers to bypass building complex boilerplate code and focus entirely on agent behavior.

A major highlight of version 0.8 is the introduction of identity-based authentication and user-level memory. While the platform previously supported durable agent-level memory, it now adds a second layer of memory scoped directly to the authenticated caller. Backed by the LangSmith Context Hub, agent-level memory is mounted at /memories/agent/ and shared across the deployment, while user-level memory is mounted at /memories/user/ and keyed to individual identities. This separation prevents personal context from leaking into shared team spaces. To secure these interactions, the update supports user-owned credentials for 23 out-of-the-box services, including GitHub, Linear, and Google Workspace, allowing agents to act with user-specific permissions.

The update also expands how users can interact with agents. In addition to existing Slack integration, Managed Deep Agents now supports Slack file transfers, allowing users to share documents, spreadsheets, and logs directly within a conversation. For external integrations, new HTTP channels allow teams to connect agents to customer portals, internal tools, or any system capable of sending webhooks. Furthermore, the release introduces built-in web search powered by Parallel. Developers can configure this tool via Model Context Protocol (MCP) servers in their project files, and LangChain is offering the Parallel search tool for free while Managed Deep Agents remains in beta.

For practitioners, these updates eliminate the need to spend quarters of engineering roadmap time building custom infrastructure for production agents. Developers can initialize a project using the command uvx -- from managed-deepagents mda init my-agent and deploy it using uv run mda deploy. This code-first approach organizes dependencies, prompt instructions, identity configurations, and memory declarations into a single directory, streamlining the path from prototype to production.