セルフホスト版 Claude Managed Agents 提供開始|サービス全体を自社インフラで運用 #AI - Qiita
@picnic

Self-hosted Claude Managed Agents release | Operate the entire service on your own infrastructure

  • Claude
  • Anthropic
  • AI
  • LLM
  • API
0
Posted at 2026-11-12

Introduction

Anthropic announced the release of the self-hosted version of Claude Managed Agents in the November 12, 2026 release notes.

Previously, the self-hosted sandbox only allowed running the sandbox portion of the agents in-house. With this self-hosted version, you can operate the entire Managed Agents service on your own infrastructure. Furthermore, agents, sessions, and session history are also stored on your side.

Why is this important? For organizations with strict requirements on data storage locations and processing paths (such as financial, healthcare, public sector, and large enterprises with strict internal regulations), the barrier of "wanting to use Managed Agents but unable to store session history externally" may be lowered.

📌 People affected

  • Information systems department and security teams who want to keep session history and agent definitions under their own management
  • Teams that previously used the traditional self-hosted sandbox but want to bring session storage in-house
  • Projects that wanted to deploy Managed Agents but were held back by data location and audit requirements

Note that applying to Anthropic's account team is necessary for usage. Please be aware that this is not a feature that can be activated immediately for anyone.

Overview of changes

We will illustrate the differences between the traditional self-hosted sandbox and this new self-hosted version in a diagram.

The flow of requests is as follows.

Changes

Comparison with the past

Item Traditional self-hosted sandbox Self-hosted Managed Agents
Scope of in-house operation Only the sandbox portion Entire service
Agent storage location (Except sandbox) service side In-house infrastructure
Session - Session history storage location (Except sandbox) service side In-house infrastructure
Model request destination Not specified in provided data Provider chosen by user (Default: Anthropic)
Conditions for use Not specified in provided data Application to Anthropic account team

💡 Tips
The items marked "Not specified in provided data" in the table are parts that could not be confirmed from the current announcement. For details, we recommend checking the official documentation and consulting with the account team.

Points summary

  • Full in-house operation of the service: The entire managed agents mechanism, not just sandbox, runs on in-house infrastructure.
  • In-house data storage: Agents, sessions, and session history are stored on in-house infrastructure.
  • Provider selection: Model requests from the server are sent to the provider chosen by the user. If no provider is selected, the default is Anthropic.
  • Application required: Application to the account team is required for use.

Impact and response

This change has action_required: false, meaning it is not a change that existing users must necessarily take action on. Existing use of Managed Agents or self-hosted sandbox can continue as is. However, it may be worth considering based on requirements.

Actions to take

  1. Requirement inventory: From data storage location, audit, and regulatory constraints, organize whether full in-house operation is truly necessary.
  2. Operational load estimation: When operating the entire service in-house, operational responsibilities such as availability, monitoring, and updates will fall on the company side. The scope will be broader than sandbox-only operation.
  3. Application to account team: Application is required for use. Consulting early will facilitate the consideration process.
  4. Model request destination confirmation: The default is Anthropic, but it is also necessary to confirm the data handling of the selected provider.
  5. Internal Document Update: Adding the self-hosted version as a deployment option in the development guide or CLAUDE.md, and clearly stating the differences from the previous self-hosted sandbox, can help prevent confusion.

⚠️ Note
Since "model requests are sent to the chosen provider," even if session history is stored in-house, the content of the prompt will be transmitted to the provider. When designing data boundaries, please consider the storage and transmission destinations separately.

Code Example

The content of this presentation does not include setup procedures or API changes. Therefore, we cannot provide specific Before/After code examples. As a reference for configuration ideas, we show an example of additions to CLAUDE.md based on the provided update proposals.

Before

## Claude Managed Agents / Deployment Options - Use Anthropic-hosted version - Option to operate only the sandbox in-house (self-hosted sandbox)

After

## Claude Managed Agents / Deployment Options - Use Anthropic-hosted version - Self-hosted sandbox: Operate only the sandbox in-house - Self-hosted version of Managed Agents (launched on November 12, 2026): - Operate the entire service using in-house infrastructure - Storage location for agents, sessions, and session history is in-house infrastructure - Model provider can be selected (default is Anthropic) - Access requires application to the Anthropic account team

Summary

  • On November 12, 2026, the self-hosted version of Claude Managed Agents was launched. Self-hosted Claude Managed Agents
  • Unlike the previous setup where only the sandbox was operated in-house, you can now place the entire service, agents, sessions, and session history on your in-house infrastructure. The destination for model requests can be selected by the user, with the default being Anthropic.
  • Access requires an application to the Anthropic account team.
  • There is no mandatory action required for existing users, but organizations that had previously postponed implementation due to data location or audit requirements may want to consider this.
  • On the other hand, since the operational responsibility of the entire service shifts to the company, system setup and designing data boundaries that separate storage and transmission destinations are important.

For detailed procedures and conditions, please check the official release notes and guidance from the account team.

0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do by signing up
Sign upLogin