Anthropic launched a self-hosted version of its Claude managed agents service on Thursday, allowing businesses to run the platform inside their own Cloud account or data centre. The Next Web reports that this new setup keeps session histories and credentials directly on internal systems.
Key takeaways
- Anthropic released the self-hosted option on Thursday for its tool-using AI agents.
- Agent histories and credentials remain on company systems rather than external cloud storage.
- Every model call still travels to an outside provider like Anthropic, Amazon Bedrock, Google Cloud, or Microsoft Foundry.
- Customers must handle server security and updates while rebuilding saved data from scratch.
How do Claude managed agents work on local servers?
Managed Agents is Anthropic's service for running AI agents that use tools and remember past sessions. Previously, customers could run agents' tools on their own machines while Anthropic managed the rest. Now, according to The Next Web, the whole service operates within a company Cloud account or data centre.
However, Claude itself does not run inside the local network. Each time the agent asks a question, the conversation goes to a model provider. Customers can choose between Anthropic, Amazon Bedrock, Google Cloud, or Microsoft Foundry.
What are the limits of this Anthropic service?
Adopting the self-hosted option means taking on new responsibilities. Customers must secure and update the server themselves. Nothing transfers over from Anthropic's hosted service, meaning saved data must be rebuilt. Certain features like webhooks and scheduled runs are also missing from this version.
Why it matters
The move gives cautious companies more control over where agent data lives. The Next Web points out that this matters in Europe, where European Central Bank president Christine Lagarde warned about relying on foreign AI. Rivals such as Mistral already pitch sovereign AI, while OpenAI took a different route in September by letting companies run agents inside Amazon's cloud.
Frequently asked questions
The self-hosted version of Claude Managed Agents launched on Thursday, according to the release notes.
Agent histories and credentials stay on company systems inside their own Cloud account or data centre.
No, Claude itself does not run inside the local network. Each time the agent asks a question, the conversation goes to an outside provider.
Customers can choose model providers such as Anthropic, Amazon Bedrock, Google Cloud, or Microsoft Foundry.
Features such as webhooks and scheduled runs are currently missing from the self-hosted setup.
Based on reporting by The Next Web. Read the original article
