Vibeleaderboard
← All Intel
Intel / post

Claude Managed Agents add sandbox memory, domain controls, session viewer

Source
x.com
Date
ClaudeDevs@ClaudeDevs

We shipped three updates to Claude Managed Agents. First, we made it easier to use memory with Self-Hosted Sandboxes. Any work done in a Self-Hosted Sandbox can be saved to memory.

Context

Agents built on Claude Managed Agents can now keep long-term memory even when their tools run on a company's own machines. Managed Agents is Anthropic's hosted service for running agents: Claude runs on Anthropic's side and requests tool calls, and a self-hosted moves the execution of those calls out of Anthropic's cloud and into infrastructure the customer controls, according to the self-hosted sandbox documentation. Memory stores are the feature that lets an carry preferences, project conventions and past mistakes from one session to the next instead of starting fresh each time. Anthropic's August 19 release notes say self-hosted sessions can now attach those stores.

The setup guide for memory in self-hosted sandboxes shows what changes for the operator. The worker process the customer runs downloads each store into a directory under /mnt/memory, the agent edits it with its ordinary file tools, and the worker syncs changes in both directions at most every 15 seconds by default, with a final flush when the session ends. Anthropic's copy stays the source of truth. If a memory changed both in the store and locally since the last sync, the store's version wins and the agent's edit is overwritten without an error reaching the agent. A change made in one self-hosted session therefore reaches another running session only after both have synced, typically under a minute, whereas cloud sessions see each other's changes almost immediately. A worker that is killed rather than stopped loses its unsynced edits. Windows hosts are not supported, and the guide says self-hosted sessions cannot attach memory stores on Claude Platform on AWS or on Anthropic-managed Google Cloud.

The post opens a thread announcing three Managed Agents updates. The same day's release notes list the other two: developers can give the web search and web fetch tools lists of allowed or blocked domains, and the Console's session viewer was redesigned with a timeline, a transcript grouped by model request, and an inspector showing cost and per-tool statistics. The memory feature itself is still labeled beta in the memory documentation.

Terms in this piece · Glossary
  • sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
More from ClaudeDevs
Recommended reads
Comments

Checking sign-in…

Loading comments…