Running an AI agent on ordinary cloud servers leaves developers rebuilding the same scaffolding: somewhere to keep context alive, somewhere for artifacts to survive the agent that made them, and a safe way to hand over tools. DigitalOcean’s Managed Agents, now in public preview, aims to supply all three.
Two pieces do the work. A Harness Runtime hands each session its own isolated microVM. The supported harnesses run from coding agents like Claude Code, Codex CLI, and OpenCode to general-purpose assistants such as Hermes, plus custom builds on LangGraph. Context and working state survive between sessions, which can be paused, picked back up, or forked. Idle agents stop on their own, and sessions can fan out across repositories and tasks for divide-and-conquer or map-reduce work.
An Action Gateway handles the other half, exposing a unified MCP endpoint that reaches more than 16,000 tools. That roster spans web search, web fetch, browser automation, DigitalOcean’s own infrastructure APIs, and connectors for GitHub, HubSpot, and Stripe.
The company frames the gap plainly: teams running agents on conventional virtual machines are forced to invest in plumbing work to keep context, persist artifacts, coordinate parallel work, and secure tool access. Managed Agents moves that burden onto DigitalOcean’s side of the line.
