Managed Agents
A managed agent execution framework provided by Anthropic that allows developers to manage and run automated AI agent sessions in the cloud. It supports session state management, error handling, and tool invocation within agent workflows.
Core Facts
Managed Agents refers to agents whose lifecycle is managed by the platform; developers can configure persistent memory stores, permission policies, and cross-session context without managing state storage themselves.
Workflow is deterministic: developers pre-define the node sequence and data flow paths, making the execution process predictable and traceable.
Timeline (last 90 days)
Anthropic's Managed Agents has launched a public beta of Dynamic Workflows
The managed-agent approach reduces cold-start costs and naturally fits cloud-based elastic scaling scenarios, at the cost of some control over underlying execution details.
v1.13.0 added three capabilities to Managed Agents: workflows, multiagent configuration, and thread status filtering.
v1.12.0 adds a url_sources parameter to the web_fetch tool configuration for hosted Agents.
The new version supports mounting public GitHub repositories in Managed Agents sessions without requiring an authorization_token.
v1.6.0 added auto mode tool permissions for Managed Agents, allowing managed agents to manage tool call permissions in automatic mode.
Modules such as the Dreams API, User Profiles API, and Managed Agents are not part of Claude's core conversation API, but are components of the higher-level Agent infrastructure that Anthropic is building.
Managed Agents refers to agents whose lifecycle is managed by the platform; developers can configure persistent memory stores, permission policies, and cross-session context without managing state storage themselves.
The evaluated_permission enum in Managed Agents-related events now shares a single definition across multiple events.
The event list filter for Managed Agents added typed event type values.
2 more timeline events
All Facts (14)
Managed Agents refers to agents whose lifecycle is managed by the platform; developers can configure persistent memory stores, permission policies, and cross-session context without managing state storage themselves.
75%VerifiedWorkflow is deterministic: developers pre-define the node sequence and data flow paths, making the execution process predictable and traceable.
65%UnverifiedManaged Agents启动的Agent能在自己隔离的Linux环境中执行代码
90%UnverifiedThe managed-agent approach reduces cold-start costs and naturally fits cloud-based elastic scaling scenarios, at the cost of some control over underlying execution details.
50%Unverifiedv1.13.0 added three capabilities to Managed Agents: workflows, multiagent configuration, and thread status filtering.
50%UnverifiedModules such as the Dreams API, User Profiles API, and Managed Agents are not part of Claude's core conversation API, but are components of the higher-level Agent infrastructure that Anthropic is building.
50%UnverifiedThe evaluated_permission enum in Managed Agents-related events now shares a single definition across multiple events.
50%UnverifiedThe event list filter for Managed Agents added typed event type values.
50%Unverifiedv1.6.0 added auto mode tool permissions for Managed Agents, allowing managed agents to manage tool call permissions in automatic mode.
70%UnverifiedAnthropic's Managed Agents has launched a public beta of Dynamic Workflows
50%Unverifiedv1.12.0 adds a url_sources parameter to the web_fetch tool configuration for hosted Agents.
50%UnverifiedThe new version supports mounting public GitHub repositories in Managed Agents sessions without requiring an authorization_token.
50%Unverifiedv1.10.0 fixes the session tool runner so that it stops after any end-of-turn idle (issue #969).
50%Unverifiedv1.10.0 adds a rejection stop reason and a stop_details field to session idle events for Managed Agents.
50%