managed-agentslisted

# Managed Agents ## When to activate Building applications where agents need to run autonomously in the cloud, or when the user mentions Claude Managed Agents, long-running agent tasks, or building agent-powered products via the Anthropic API. ## When NOT to use - Claude Code subagents running in a terminal session — those use the `Task` tool, not this API - Short synchronous requests that complete in under 10 seconds — use the standard Messages API - Workflows requiring Zero Data Retention (ZDR) or HIPAA BAA — Managed Agents are not eligible ## Instructions ### Core Concepts - **Agent:** a configured entity with a model, system prompt, and allowed tool set - **Environment:** a compute sandbox where the agent runs (cloud-hosted by Anthropic, or self-hosted) - **Session:** one execution run — has a start, an end, and an event stream - **Events:** Server-Sent Events (SSE) stream reporting what the agent is doing in real time **Key distinction from Claude Code subagents:** Managed Agents run independently of your terminal in Anthropic's cloud. Use them for async, long-running, or API-driven agent products — not for Claude Code slash commands. ### Beta Header All Managed Agents API calls require: ``` anthropic-beta: managed-agents-2026-04-01 ``` ### Tool Type To give an agent access to all built-in tools (Bash, file operations, web search, web fetch, MCP): ```python tools=[{"type": "agent_toolset_20260401"}] ``` ### Python Pattern ```python import anthropic client = anth