
Anthropic 现允许企业在自己的云环境或数据中心运行 Claude Managed Agents。代理历史记录和凭证保留在其系统内,但每次模型调用仍会发送给外部提供商,如 Anthropic 或 AWS。
Anthropic now lets companies run Claude Managed Agents in their own cloud or data centre. Agent histories and credentials stay on their systems, but each model call still goes to an outside provider such as Anthropic or AWS.
据 Help Net Security 报道,Anthropic 现允许企业在自己的服务器上运行其 Claude Managed Agents 服务。该公司在发布说明中表示,自托管版本于周四推出,客户必须通过其 Anthropic 账户团队申请访问权限。
Anthropic now lets companies run its Claude Managed Agents service on their own servers, Help Net Security reported. The self-hosted version launched on Thursday, and customers must request access through their Anthropic account team, the company said in its release notes.
Managed Agents 是 Anthropic 用于运行 AI 代理的服务,这些代理能够使用工具并记忆过去的会话。客户此前已可在自己的机器上运行代理的工具,但 Anthropic 负责运行其余部分,包括存储的历史记录和凭证。
Managed Agents is Anthropic’s service for running AI agents that use tools and remember past sessions. Customers could already run the agents’ tools on their own machines, but Anthropic ran the rest, including the stored history and credentials.
现在客户可以在自己的云账户或数据中心运行整个服务。代理、会话历史记录和凭证保留在其系统上,且每次工具调用都可被记录。
Now customers can run the whole service in their own cloud account or data centre. Agents, session histories and credentials stay on their systems, and every tool call can be logged.
但有一个限制。每当代理向 Claude 提问时,对话仍会发送给模型提供商,如 Anthropic、Amazon Bedrock、Google Cloud 或 Microsoft Foundry。客户可自行选择提供商,但 Claude 本身并不在其网络内运行。
There is a catch. Each time the agent asks Claude a question, the conversation still goes to a model provider, such as Anthropic, Amazon Bedrock, Google Cloud or Microsoft Foundry. Customers choose which one, but Claude itself does not run inside their network.
客户还需承担 Anthropic 过去负责的工作,如保护和更新服务器。无任何内容从 Anthropic 的托管服务迁移过来,因此代理和保存的数据必须重建,且部分功能(如 webhook 和定时运行)缺失。
Customers also take on the work Anthropic used to do, such as securing and updating the server. Nothing moves over from Anthropic’s hosted service, so agents and saved data must be rebuilt, and some features, such as webhooks and scheduled runs, are missing.
这一举措让谨慎的企业能更好地控制代理数据的存放位置。这在欧洲尤为重要,欧洲央行行长拉加德曾警告依赖外国 AI 的风险。竞争对手如 Mistral 长期以来一直将主权 AI 作为答案进行推广。
The move gives cautious companies more control over where agent data lives. That matters in Europe, where ECB president Christine Lagarde has warned of the risk of relying on foreign AI. Rivals such as Mistral have long pitched sovereign AI as the answer.
OpenAI 9 月采取了不同路径,允许企业在亚马逊云中运行其代理。与此同时,英伟达已为更安全的代理构建了硬件。
OpenAI took a different route in September, letting companies run its agents inside Amazon’s cloud. Nvidia, meanwhile, has built hardware for safer agents.