Originally reported November 16, 2026.
DigitalOcean Managed Agents was in public preview, offering a managed cloud environment for AI agents. The service combines Harness Runtime for agent execution with Action Gateway for access to external tools. DigitalOcean says the aim is to reduce the work developers would otherwise do to maintain agent state, provision compute and secure tool access.
Harness Runtime uses isolated microVMs to run supported agents, including Claude Code, Codex CLI, OpenCode and Hermes, as well as custom agents built with LangGraph. According to the report, sessions can retain conversation history and work state, then be paused, resumed or forked. Idle sessions pause automatically when no LLM or tool call is in progress. The runtime also supports parallel sessions and connections to internal services without exposing those services to the public internet.
Action Gateway provides a unified MCP endpoint for more than 16,000 tools, according to the report. Examples include web search, browser automation, DigitalOcean infrastructure APIs and connectors for GitHub, HubSpot and Stripe. Its controls include per-agent permissions, mandatory human approval for sensitive actions, and rate limiting, retries, backoff and timeouts.
The report compares Managed Agents with Docker Cloud Sandboxes, noting that both offer microVM isolation and persistence. It characterizes Docker’s offering as more focused on developer workflows and movement between local and cloud compute, while DigitalOcean’s emphasizes managed infrastructure and tool access for production-oriented agent workloads.
Source: infoq.cn (originally published 2026-11-16 16:42 UTC+8)


