Module managed_agents 

Available on crate feature managed-agents-preview only.

Anthropic Managed Agents API (preview).

Managed Agents lets you provision long-running agent sessions backed by Anthropic-managed compute environments, persistent memory stores, and credential vaults. Each session references a versioned agent and an environment, drives a stream of events, and may produce outputs as files.

Beta status

All requests in this module require the managed-agents-2026-04-01 beta header, which the SDK adds automatically. Outcomes additionally require managed-agents-2026-04-01-research-preview.

The API surface is in preview and will change. Field names, request shapes, and resource relationships are not yet stable. We follow the broader claude-api forward-compatibility contract: every union deserializes into Other(Value) when the wire type tag is unknown, so brand-new variants don’t break the build. New Known variants are minor bumps that may require sweeping Other matches.

Gated on the managed-agents-preview feature.

Layout

• sessions – create, retrieve, list, archive, delete sessions; send and stream events.
• vaults – credential vaults for MCP authentication.
• memory_stores – persistent memory across sessions.
• agents – agent definitions (create only in this version; full CRUD lands once docs are available).

Modules

agents

Agents: full CRUD + version history.

environments

Environments: container configuration for sessions.

events

Session events: typed user / agent / session / span events.

memory_stores

Memory stores: persistent text documents that survive across sessions.

resources

Session resources: file / github_repository / memory_store mounts.

sessions

Sessions: provision, retrieve, list, archive, delete.

threads

Multi-agent session threads.

vaults

Vaults: per-user collections of MCP credentials.

Structs

ManagedAgents

Top-level namespace handle for the Managed Agents API.