Anthropic lets companies run Claude Managed Agents on their own servers in a private beta
Customers host the agent harness, sandboxes, credentials and session history in their own cloud or on-premises and choose where model requests go, though Claude itself still runs outside their network.
Why it matters
Until now Anthropic ran everything in Managed Agents except, optionally, the sandbox; teams wanting the whole service inside their perimeter had to build it themselves. Self-hosting targets agents that work unattended on company code, such as fixing builds or triaging alerts, where credentials and audit trails are the main security worry. Model calls can go to Anthropic, Amazon Bedrock, Google's Agent Platform, Microsoft Foundry or a customer gateway. Pricing, eligibility and when the beta opens more widely are still to be announced.
Key facts
- Private beta opened on 12 November 2026
- Access is requested through an Anthropic account team
- Customer runs the Managed Agents server, its PostgreSQL database and the workers
- Requests can go to Anthropic, Amazon Bedrock, Google's Agent Platform, Microsoft Foundry or a gateway
- Claude itself still runs outside the customer's network