Long-running agent workflows
Applications that need to pause, resume, recover and preserve a detailed history across multi-step work.
Independent tool overview
Claude Managed Agents at a glance
Claude Managed Agents is Anthropic's managed runtime and API layer for building persistent, tool-using Claude agents with sandboxes, memory, observability and multi-agent orchestration.
Visit the official Claude Managed Agents site ↗
- Best for
- production Claude agents that need managed state and execution
- Availability
- public beta on the Claude Platform
- Runtime
- Anthropic-managed or self-hosted sandbox environments
- Pricing
- Claude tokens plus $0.08 per active session-hour
Overview
What Claude Managed Agents is
Claude Managed Agents packages the operational layer around a Claude-powered agent. Developers define the model, prompt, tools and guardrails; Anthropic handles the agent loop, persistent event stream, execution environment, retries and session lifecycle.
The product is organized around agents, environments and sessions. A session can run for an extended period, pause while idle, resume with its event history and sandbox state intact and expose a trace of model and tool activity in the Claude Console.
Managed Agents entered public beta in April 2026. Memory, outcomes and multi-agent orchestration are also beta capabilities, while Dreaming—the scheduled process that curates patterns from sessions and memory—is a research preview requiring access.
Use cases
Who Claude Managed Agents is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Tool-using production agents
Teams that want sandbox execution, credentials, permissions, webhooks and traces without maintaining the entire agent harness.
Parallel specialist systems
Complex tasks where a lead agent can delegate independent work to context-isolated specialists running concurrently.
Capabilities
Core Claude Managed Agents features
Persistent sessions
Stores events, outputs and sandbox state so work can stream, idle, resume and be inspected after execution.
Managed and self-hosted environments
Runs tools in Anthropic cloud sandboxes by default or connects self-hosted workers and private MCP servers for tighter infrastructure control.
Memory stores
Mounts exportable, versioned files that agents can read and update across sessions with scoped access and audit history.
Outcomes and grading
Lets developers define a success rubric that a separate grader uses to evaluate the agent's output.
Multi-agent orchestration
Allows a coordinator to delegate parallel tasks to agents with their own model, prompt, tools and isolated context threads.
Dreaming
Research-preview scheduled processing that reviews sessions and memory, surfaces recurring patterns and proposes or applies memory updates.
Process
How the Claude Managed Agents workflow works
Step 1
Define one bounded agent
Start with a clear task, tool allowlist, permissions, success rubric and escalation path before adding multiple agents.
Step 2
Choose the execution boundary
Decide whether Anthropic-hosted containers, a self-hosted sandbox or private MCP tunnels fit the data and tool risk.
Step 3
Test traces and failure recovery
Evaluate real event timelines, retries, cost, memory changes and human approval points before production traffic.
Cost
Claude Managed Agents pricing and free plan
Claude Managed Agents uses consumption pricing rather than a separate seat plan. Customers pay standard Claude model token rates plus $0.08 for each hour a session is actively running. Idle, rescheduling and terminated time is not billed as runtime; optional web search and external services can add usage charges.
Claude model usage
Standard model token rates
Input, output and cache tokens consumed by every agent and grader call.
- Rate depends on the selected Claude model
- Prompt caching rates apply
- Multi-agent and outcome-grading calls add their own token usage
- No Batch API discount for stateful Managed Agent sessions
Session runtime
$0.08/session-hour
Metered runtime while a session is actively running.
- Measured to the millisecond
- Idle, rescheduling and terminated states do not accrue runtime
- Replaces separate code-execution container-hour billing
Web search
$10 per 1,000 searches
Optional search calls triggered inside a Managed Agent session.
- Charged in addition to tokens and runtime
- Tool and third-party API costs may be separate
- Use permissions and budgets to control unbounded tool usage
Pricing checked . Check current pricing at the source ↗
Assessment
Claude Managed Agents strengths and limitations
Where it stands out
- Reduces the infrastructure work required for stateful Claude agents, including sandboxes, event persistence and observability.
- Self-hosted execution and private MCP options provide more deployment flexibility than a cloud-only agent runtime.
- Versioned memory and traceable multi-agent events make long-running behavior easier to inspect and manage.
What to consider
- Managed Agents and several headline capabilities remain beta or research-preview features and may change before general availability.
- The harness is optimized for Claude, so adopting it creates more model and platform coupling than a model-neutral orchestration layer.
- Parallel agents, graders, memory processing and long sessions can multiply token and tool costs if budgets are not enforced.
- Persistent memory and self-modification require explicit review, redaction, retention and rollback policies for sensitive workloads.
Compare
Claude Managed Agents alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
LangChain
Choose LangChain when a model-neutral framework and more control over orchestration infrastructure matter more than Anthropic's managed Claude runtime.
Explore LangChain →Questions
Claude Managed Agents FAQs
What is Claude Managed Agents?
It is Anthropic's API and managed runtime for building tool-using Claude agents with persistent sessions, sandboxes, permissions, memory, observability and orchestration.
Is Claude Managed Agents generally available?
No. The core product is in public beta as of this review. Dreaming is a research preview, and access or behavior can change.
How much does Claude Managed Agents cost?
You pay the selected Claude model's normal token rates plus $0.08 per actively running session-hour. Web search costs $10 per 1,000 searches, and other tools can have separate charges.
Can Managed Agents run inside my infrastructure?
Yes. Anthropic supports self-hosted sandbox workers and MCP tunnels for private MCP servers, while session control and model inference still use the Claude Platform.
What is Dreaming in Claude Managed Agents?
Dreaming is a scheduled research-preview process that reviews prior sessions and memory stores, extracts patterns and curates memory automatically or with human review.
Bottom line
Our Claude Managed Agents verdict
Claude Managed Agents is attractive for teams that have already selected Claude and do not want to build the state, sandbox, trace and memory layers around it from scratch. The tradeoff is stronger platform coupling and exposure to beta changes, so production adoption should start with bounded tasks, hard budgets and explicit approval controls for tools and memory.
Visit Claude Managed Agents website ↗Similar Agents Tools
View all Agents AI tools →
Codex Pets
Codex Pets - OpenAI's animated companions for tracking active Codex work from a floating overlay
Higgsfield Supercomputer
Higgsfield Supercomputer - ACloud-based AI agent with built-in tools, persistent memory, and scheduled task automation
Link for Agents
Link - Stripe's new payment wallet for AI bots with human approval on every purchase
Gemini Spark
Gemini Spark - Google's personal agent that runs 24/7 on Cloud VMs
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.