DigitalOcean Holdings, Inc. (NYSE: DOCN) announced the public preview of DigitalOcean Managed Agents, a set of services that provides isolated runtime environments, tool access, and serverless inference for AI agent workloads on a single platform.

According to the company, Managed Agents is built from two services — Harness Runtime and Action Gateway — integrated with DigitalOcean's Inference Engine. Harness Runtime provides hardware-isolated microVM environments with lifecycle APIs for pausing, resuming, and forking agent sessions. Action Gateway gives agents access to more than 16,000 tools from over 500 providers through a single managed endpoint, with credentials brokered at execution time. The Inference Engine connects agents to more than 75 open and proprietary models.

Pricing is structured around active CPU usage at $0.044 per vCPU-hour and memory at $0.0095 per GB-hour, with CPU charges paused when agents are waiting on model responses or tool results. Snapshots are priced at $0.005 per GiB-month. DigitalOcean claims this pricing model can result in up to 37% lower monthly total cost of ownership compared to a leading independent sandbox provider, based on its own testing.

The company states that Harness Runtimes resume from a paused state in 305 milliseconds, which it claims is 46% faster than other leading offerings, and completes end-to-end microVM creation to agent response 31% faster, based on internal testing.

Managed Agents supports coding harnesses including Claude Code, Codex CLI, and OpenCode, as well as general-purpose agents such as Hermes and agents built with LangGraph. Custom agents packaged as Open Container Initiative container images are also supported.

OpenHands, Qencode, and Amplitude are among the companies that participated in the private preview and are building on the platform. Qencode reported using Managed Agents to build a support-triage agent that the company estimates saves its team four to eight hours per week.

New DigitalOcean users receive a $5 credit to launch their first agent session. The service is currently offered as a public preview, and DigitalOcean notes it does not guarantee that preview services perform for production-level workloads.