Anthropic opens self-hosted Managed Agents private beta

Anthropic says Managed Agents provides composable APIs for building and deploying autonomous agents, along with an agent harness, sandboxes, credential handling and session history.[1] The self-hosted offering runs every part of that service in the customer’s own cloud or on-premises, including the harness, sandboxes, credentials and session history.[1] Developers can build agents that work directly in company code and internal systems.[1] Anthropic says the agents can run unattended on customer code and need access to internal systems.[1]

Customers operate the Managed Agents server, its database and the workers where agents run shell and file tools, Anthropic says.[1] Its quickstart describes a one-machine setup using Docker Compose, with a server, a PostgreSQL database and workers; each session runs in its own sandbox container.[2] Anthropic says the self-hosted server has the same API as its hosted Managed Agents service, subject to differences listed in the documentation.[3] Teams considering a deployment should review those differences and plan for operating the listed components themselves.

Anthropic first launched Managed Agents as a hosted service on Claude Platform, with Anthropic running the infrastructure.[1] It later added self-hosted sandboxes for teams that needed code to remain in their own environment.[1] Anthropic says some teams wanted the whole service in their environment, rather than only the sandbox, and previously had to build the platform themselves.[1] Its examples of agent work include fixing broken builds, testing releases, triaging alerts and working through a bug backlog.[1]