Voyage Digital
    Tools· AI Assistance & Agent Building

    Claude Managed Agents

    Anthropic's managed cloud infrastructure for running Claude as an autonomous, tool-using agent without building your own runtime.

    automatingexecutablekommerziell
    Open provider
    Claude Managed Agents

    Description

    Instead of running your own agent loop, tool execution, and runtime environment, with Claude Managed Agents you define the model, system prompt, tools, MCP servers, and skills once and then reference the agent by an ID in further sessions. The agent reads files, executes commands, searches the web, and runs code in a secured, managed environment, including built-in prompt caching and compaction for more efficient runs.

    The offering is designed for long running tasks spanning several minutes or hours with many tool calls, as well as for recurring, scheduled runs through planned deployments. Communication runs through events, and responses come back via server sent events.

    As of the source, Claude Managed Agents is in beta status and is accessed through a separate beta header in the API. It is aimed at teams that want to run their own Claude based agents in production, not at individual users in the regular Claude chat or in Claude Code, for whom the usual access is sufficient.

    Strengths

    Managed runtime
    the agent loop, tool execution, and execution environment run at Anthropic, so no infrastructure of your own is needed
    Prompt caching and compaction
    built in for more efficient runs during long, tool heavy sessions
    Stateful, resumable sessions
    conversation history, sandbox state, and outputs are kept server side, sessions can pause and resume cleanly
    Steering mid execution
    additional events can be sent partway through a run to guide the agent or interrupt it

    Assessment

    AI features

    • Agent definition set the model, system prompt, tools, MCP servers, and skills once, then reference the agent by its ID
    • Environment configuration determines where the agent runs, including in a cloud sandbox
    • Sessions a session ties together the agent and environment configuration and starts the actual run
    • Event streaming user messages go in as events, Claude runs tools autonomously and streams results back via server sent events, with the full history kept server side
    • Steer or interrupt send additional user events during execution to change direction

    Suitable for

    • Developer teams that want to run their own Claude based agents in production on an ongoing basis, without building their own runtime environment
    • Use cases with long running tasks spanning several minutes or hours with many tool calls
    • Recurring, scheduled agent runs that need to execute regularly without a manual start
    • Less suitable for Less suitable for simple, interactive one off requests, where the regular Claude access is sufficient

    Limitations and notes

    • Beta status according to the source, Claude Managed Agents is in beta and access runs through a separate beta header in the API
    • Target audience built for developer teams running agents in production, for simple one off requests in the regular Claude chat or in Claude Code the usual access is sufficient
    • Server side data because of the stateful design, conversation history, sandbox state, and outputs are stored server side
    • AWS variant differs according to the source, also available on Claude Platform on AWS, where feature availability and session behavior differ in some respects

    Quick start

    1. Check access: an Anthropic API account is enough, Managed Agents is enabled there by default according to the source, and calls go through the Claude Platform API with a separate beta header
    2. Define the agent: set the model, system prompt, tools, MCP servers, and skills, and create the agent once
    3. Configure the environment: decide where the agent runs, for example in a cloud sandbox
    4. Start a session: create a session that references the agent and environment configuration
    5. Send the first event: send a user message as an event and stream the response back via server sent events
    6. Steer when needed: send further events during execution to guide the agent, or interrupt it

    Tips

    • Use the event history for traceability, the full history is kept server side and can be retrieved in full
    • When deploying on AWS, check the differences in feature availability and session behavior beforehand
    • For recurring tasks, use planned deployments instead of manual starts
    • Before going to production, check the current state of the documentation, since the offering is in beta

    Access

    According to the source, Claude Managed Agents is in beta status and is accessed through the Claude platform API with a separate beta header. Billing follows Anthropic's usual API usage model; the source does not name separate rates specifically for the managed infrastructure.

    Last reviewed: · Pricing, plans and features are a snapshot in time. Check the provider's own page before deciding.

    In the workshop this becomes your method.

    Whoever sees this process run once wants the agent behind it next. We build that in the workshop From Process to Agent.

    View workshops

    Related resources

    Browse all resources
    Agentforce
    Tools

    Agentforce

    Salesforce's platform for autonomous AI agents in sales, support, marketing and commerce, deeply integrated into the Salesforce Data Cloud.

    Learn more
    Copilot 365
    Tools

    Copilot 365

    Microsoft's AI assistant, integrated directly into Word, Excel, PowerPoint, Outlook and Teams, designed for companies already working within the Microsoft 365 ecosystem.

    Learn more
    EV
    Tools

    everything-claude-code (ECC)

    everything-claude-code (ECC) is an MIT-licensed meta-collection of skills, agents, slash commands, hooks and rules packs that standardizes agentic coding workflows across multiple AI coding harnesses.

    Learn more
    Langdock
    Tools

    Langdock

    Enterprise AI platform from Berlin that gives teams central, governed access to multiple language models, Custom Assistants and no-code agents under GDPR compliant EU hosting.

    Learn more
    Lindy
    Tools

    Lindy

    Lindy is a no-code platform for building AI agents that handle email, meetings, lead qualification and customer support for you.

    Learn more
    Relevance AI
    Tools

    Relevance AI

    Relevance AI is a no-code platform for building autonomous AI agents for sales, support and operations, with an enterprise focus and an EU hosting option.

    Learn more
    ChatGPT
    Tools

    ChatGPT

    OpenAI's AI assistant built around GPT-5, with Canvas for collaborative editing, an Agent Mode for browser tasks, and image and video generation through DALL-E 3 and Sora.

    Learn more
    Claude
    Tools

    Claude

    Anthropic's AI assistant family (Opus, Sonnet, Haiku), used as a core tool throughout the Voyage Digital stack for development and operations, strong at coding, long documents, and agentic workflows via MCP servers.

    Learn more
    Grok
    Tools

    Grok

    Grok is xAI's AI assistant with real-time access to X, a multi-step DeepSearch research function, image generation, voice mode, and task-based research agents, particularly strong for current events and trend research.

    Learn more
    LibreChat
    Tools

    LibreChat

    LibreChat is a self-hostable open-source chat frontend that unifies multiple AI providers, agents and MCP servers in a single interface with single sign-on and user management.

    Learn more
    Open WebUI
    Tools

    Open WebUI

    Open WebUI is a self-hostable, offline-capable interface for local and remote language models, with Ollama integration, local RAG and enterprise features such as SSO and audit logs.

    Learn more
    Skills

    The Role Rulebook

    Guides a project through five clearly separated roles, Explorer, Planner, Builder, Reviewer and Navigator, so the actual problem is understood before anything gets planned, built or reviewed.

    Learn more
    Assistants

    The Assistant Curator

    Turns your briefing into a complete, ready-to-use AI assistant configuration, including system prompt, knowledge sources, tool selection and a reasoned model recommendation.

    Learn more

    Conversation, not pitch

    Understand first, then decide. We take time for an initial conversation, without sales pressure, without obligation.

    Schedule a call