Open Managed Agents
An open-source, self-hosted alternative to Claude Managed Agents and the OpenAI Agents API that runs agents with durable sessions, sandboxed tools, memory and encrypted credentials on a developer's own infrastructure.
Open Managed Agents lets a developer run the Claude Managed Agents API or the official OpenAI Agents SDK against their own deployment instead of a hosted vendor service, bringing their own model keys and choosing where the agent's data lives. It exposes the Claude-compatible /v1/agents and /v1/sessions endpoints and the OpenAI Agents API at /openai/v1 from either of two hosting modes that share the same underlying business logic and event-log model: a Node.js self-host running on a VPS, a Mac, Docker, fly.io or a Kubernetes cluster with SQLite or Postgres storage and a sandbox such as LiteBox, Daytona, E2B or BoxRun, or a Cloudflare deployment using Workers, Durable Objects and Containers with D1, KV and R2 storage.
Sessions support text and function calls that wait for the calling application's own result, and the Node host additionally supports dynamic subagents: a main agent can delegate subtasks to child sessions that keep separate conversations, share the parent's files, and cannot themselves spawn further children. A hosted version of the same software is also offered for developers who would rather not run their own deployment.