LangChain Adds Personal Memory to Managed Agents, Off by Default in Group Chats

A team’s AI agent can now remember one person’s preferences without putting them in the memory it shares with everyone. LangChain launched Managed Deep Agents 0.8 on September 24, 2026, with a memory layer tied to each authenticated caller. The release also adds Slack file transfer, HTTP channels and web search powered by Parallel.

Personal context, separate from team context

Managed Deep Agents already had durable memory for instructions and preferences shared across conversations. Version 0.8 adds a layer keyed to the person starting a run. LangChain says it can retain that person’s working style or recurring tasks without mixing those details into shared agent memory.

The channel sets the default boundary

By default, a one-to-one Slack message can use both memory layers. Slack groups and channels, and HTTP requests, can use shared agent memory but not user memory. Teams can change those access policies. The distinction keeps a person’s stored preferences out of group conversations under the default settings; it is not an unconditional ban.

Remembering someone is different from acting as them

Memory determines what context the agent can recall. Credentials determine whose permissions it uses in another service. Managed Deep Agents supports credentials owned by an individual user as well as credentials shared by the agent across callers. Those connections are distinct from the new memory layer. LangChain says they support 23 services, including GitHub, Linear and Google Workspace tools.

More ways to hand over a task

Slack users can now send files with a request, so an agent can work from a spreadsheet, screenshot or document supplied in the conversation. New HTTP channels accept JSON webhooks, a way for software to send requests to an agent. LangChain suggests internal tools, customer portals and support systems as possible places to use them.

Search is built in, but plan-limited

For web search, LangSmith manages the Parallel credentials and runs the tool. Teams do not need a separate Parallel account or API key. LangChain says the agent receives excerpts and source URLs, while search calls, delays and errors appear in LangSmith traces. Its documentation describes the integration as offering zero data retention.

Built-in search is available only on LangSmith Plus and Startup plans, with usage free during private beta. Setup still requires a Managed Deep Agents project, a LangSmith workspace ID and a personal LangSmith API key for Studio. The integration removes a separate search-vendor setup step, not the LangSmith requirements.