What it does
5 features
Claude Managed Agents · Agent platform
Product and access
Developer platform in beta; a distinct offering alongside Claude Code and Cowork.
Core features
5 features from provider sources · checked on 19 October 2026. Availability and limitations follow below.
Usable AI models
Own models only
You set a Claude model per agent, and the docs list only Claude models as supported. You need a Claude API key and can also use the service through Claude Platform on AWS or Claude Models on Google Cloud. The whole offering is still in beta.
Model families named
- Anthropic (Claude)
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- No
- Local models
- No
The rating covers the generally available main access. Documented · Sources accessed on
Does it fit my task?
2 practical tasks · 3 requirements and limits
A good fit if …
- you want to build your own agents on the Claude API and can plan for an API key and development effort
- tasks should run for hours and recur on a cron schedule with run history and webhooks
- you want sandboxes on your own infrastructure for compliance or data residency reasons
Less suitable if …
- you need zero data retention or a HIPAA BAA; neither applies to Managed Agents at present
- you expect a ready-made interface without coding; access is through the API, SDKs and a beta header
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Example task 1
- Your input
- Analyze approved documents in a background task.
- Possible result
- A draft for expert review.
- What to check
- Check inputs, output and performed actions.
Example task 2
- Your input
- Prepare a recurring repository report.
- Possible result
- A draft for expert review.
- What to check
- Check inputs, output and performed actions.
Analyze approved documents in a background task.
Your possible prompt
Design a process for new requests: capture information, sort it by topic, and create a proposed response. Plan an approval step before external actions.
Adjust the task, data and permissions.
Requirements and limits
Can I use it?
With requirements
Europe & regional availability
Provider sources reviewed again. This assessment applies to the access described here.
Reviewed scope: Product access and explicitly stated exceptions
Reviewed on:
With requirements
API beta also covers supported European countries
Germany is on the Claude API country list. Managed Agents is enabled by default for API accounts and requires an API key and beta header. MCP tunnels and dreaming require separate preview access.
Region, reason and source
Region covered: Germany; other countries according to the provider list
Documented reason: The sources state access conditions; they do not show EU regulation as a cause.
Official source · platform.claude.com · Official source · platform.claude.com
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Claude Managed Agents (beta) via the Claude API with an Anthropic cloud sandbox
Sources reviewed:
Show all 3 points
Documented
Account used
Requests use a Claude API key from the Console and the beta header. For MCP services that need sign-in, a session references a vault holding stored OAuth credentials. On Claude Platform on AWS each scheduled session runs as an IAM role in your AWS account.
Evidence · Evidence · EvidenceDocumented with conditions
Admin approval
Access is enabled by default for all API accounts. Per tool you can set whether calls run without confirmation, require confirmation or are evaluated individually by the server. An inference geography set for a session is validated against the workspace's allowed geographies.
Evidence · Evidence · EvidenceDocumented with conditions
Data access & team sharing
The agent reads and writes files in its sandbox, runs commands and calls web and MCP servers as far as you enable the tools; web access can be limited to domains. Uploaded files, GitHub, memory stores and vaults are optional resources. Team sharing is not described.
Evidence · Evidence
Memory and conversation continuity
4 of 6 criteria documented
Versioned text stores attached to selected sessions and used through file tools.
Managed Agents with memory stores (beta)
Sources accessed:
Documentation by criterion
StorageDocumented
Storage and scope
Memory stores are workspace-scoped collections of text documents mounted as session directories. Changes produce immutable versions.
SelectionDocumented
Selection and timing
Agents read and write using file tools; mount guidance is automatically added to the system prompt. What is saved follows the task and agent instructions.
Edit & deleteDocumented
Inspect, edit and delete
API and Console support inspection and editing; APIs delete memories or entire stores. Historical content requires targeted redaction or deletion of the entire store.
Access boundariesDocumented
Agent and channel boundaries
Stores can be attached and shared across sessions per user, team or project; read_only/read_write controls writes.
RetrievalPartly documented
Retrieval
The agent receives the mount path, description and instructions and reads needed files. A preliminary semantic search index is not documented.
New conversationsPartly documented
Session reset and compaction
New sessions start with fresh context; attached stores outlive sessions. Dreaming consolidates memory into a new store; active-conversation compaction is separate and not explained here.
Cost and effort
Prices as of 19 October 2026
Prices and included usage
Pricing checked
USD 0.08 per running session-hour plus model usage.
- Billing depends on plan, usage and setup.
Setup
Requires an API account, agent configuration and beta header.
Operation & oversight
Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.
Deployment covered: Managed cloud sandbox in beta; optionally your own sandbox
Sources reviewed:
Editorial assessment
Run records, webhooks and pause logic are documented. You have to set up permission policies, budgets and the review of failed runs yourself.
Show all 6 points
Documented
Files and working state
Sessions keep conversation history, sandbox state and outputs server-side and resume after pauses. Transcripts persist until you delete them; zero data retention does not apply to Managed Agents.
EvidenceDocumented with conditions
Human intervention
A permission policy can require confirmation for tool calls; the built-in tools are otherwise enabled. For custom tools the session pauses until your application returns the result.
EvidenceDocumented
Detecting errors
Each trigger attempt of a deployment creates a run record that can be filtered for errors. Session progress and outcomes also arrive through the event stream and webhooks.
EvidenceDocumented
Resuming after failures
A session that reaches its budget pauses and goes idle. For unrecoverable errors such as an archived environment or vault the deployment is paused until you update it; after unpausing, missed triggers are not backfilled, and a rate-limited session start is not retried until the next scheduled time.
Evidence · Evidence
Evidence and open questions
2 sources · 1 open point
Open points
The reviewed sources contain no reliable information on these points. A gap says nothing about the product.
Requirements and limits: not yet confirmed
Sources and evidence
Provider sources reviewed; examples are editorial suggestions. No hands-on test.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
Product and provider are documented through the linked original sources.
- Operator
- Anthropic
- Organization
- Provider
- Website
- platform.claude.com
Changes to this profile 1
Added Claude Managed Agents with capabilities, access, pricing and sources.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.