AI Catchup

Claude Managed Agents Add Memory Stores, Domain Controls, and a Redesigned Console

By 3 min read

Anthropic's Claude Managed Agents updates let self-hosted sandbox sessions attach memory stores, restrict web_search and web_fetch with allowed_domains or blocked_domains, and inspect multi-agent sessions in a redesigned Claude Console viewer.

Anthropic shipped three Claude Managed Agents updates on August 19: memory stores for self-hosted sandboxes, domain controls for web tools, and a redesigned Claude Console session viewer. The approved ClaudeDevs account announced the three changes in a short thread, while Anthropic's durable platform release notes document the implementation details. (ClaudeDevs on X; Claude Platform release notes)

The combination covers three practical parts of an agent system: persistent state, controlled retrieval, and visibility into multi-agent execution.

Memory stores in self-hosted sandboxes

Claude Managed Agents sessions running in a self-hosted sandbox can now attach memory stores. Anthropic says the Python, TypeScript, and Go SDK workers download each attached store into the sandbox at its mount_path and sync the agent's changes back to the store. (Claude Platform release notes)

That gives a self-hosted execution environment a way to preserve agent state across sessions without turning the sandbox itself into the memory system. The release notes describe the worker behavior and mount path; they do not promise a particular memory-store size, retention period, or plan entitlement.

This is an extension of Anthropic's earlier self-hosted sandbox direction. The older announcement covered where tool execution runs; this update covers how a self-hosted session can carry state into and out of that environment. (Claude Managed Agents self-hosted sandboxes)

Restricting web_search and web_fetch

Anthropic added domain controls for the web tools used by Claude Managed Agents. The release notes say teams can configure allowed_domains or blocked_domains on a tool's entry in the agent_toolset_20260401 configs array. (ClaudeDevs on X; Claude Platform release notes)

The same configuration surface supports tool-specific parameters:

  • web_fetch accepts max_content_tokens.
  • web_search accepts user_location.
  • Each configs entry is identified by its name and can carry an optional type.
  • Requests that pass only name, enabled, and permission_policy continue to work.

Domain allowlists and blocklists do not make an agent's browsing automatically correct, but they give platform teams a concrete boundary around which sites the agent may search or read. That is especially useful when an unattended workflow should use an approved documentation set rather than the open web.

A new Console session viewer

The Claude Console session viewer now has a timeline minimap, a transcript grouped by model request, and an Inspector panel for session details and cost. The release notes also list raw events, per-tool statistics, mounted resources, and per-thread activity in the redesigned viewer. (ClaudeDevs on X; Claude Platform release notes)

For multi-agent sessions, those additions make it easier to answer operational questions: which agent was active, what tools ran, what resources were mounted, and where cost accumulated. The announcement says the redesign is live; the release notes document the viewer's fields but do not claim a new pricing model or a particular retention policy for session data.

What developers should take from the update

These changes are most useful together:

  1. Attach a memory store so a self-hosted session can retain state.
  2. Restrict web_search and web_fetch to the domains the workflow is allowed to use.
  3. Use the Console viewer to inspect tool activity, mounted resources, and cost across the resulting multi-agent run.

Anthropic's release notes establish the feature behavior and date. They do not establish that every organization has identical limits or that a self-hosted sandbox automatically inherits a specific security policy, so teams should check their own Console and SDK configuration before relying on those assumptions. (Claude Platform release notes)

Sources

Keep building the workspace playbook

Frequently Asked Questions

What is new about memory in Claude Managed Agents?

Sessions running in a self-hosted sandbox can now attach memory stores. The Python, TypeScript, and Go SDK workers download each store at its mount_path and sync the agent's changes back to the store.

How can Claude Managed Agents restrict web access?

Configure allowed_domains or blocked_domains on the web_search or web_fetch tool entry in the agent_toolset_20260401 configs array. web_fetch also accepts max_content_tokens, while web_search accepts user_location.

What changed in the Claude Console session viewer?

The redesigned viewer adds a timeline minimap, a transcript grouped by model request, and an Inspector panel with session details and cost, raw events, per-tool statistics, mounted resources, and per-thread activity.

Get the weekly AI Catchup

Tools, practices, and what matters, in your inbox every week.