Managed Agents | DigitalOcean

Featured AI Products
Build, deploy, and scale cloud compute resources
Containers and Images
Safely store and manage containers and backups
Fully managed resources running popular database engines
Management and Dev Tools
Control infrastructure and gather insights
Secure and control traffic to apps
Help protect your account and resources with these security features
Store and access any amount of data reliably in the cloud
Security and Networking
Web and App Platforms
One AI-native stack to power your intelligence. Deploy your preferred agent harness or bring your own, connect it safely to 16,000+ tools, and let it do more work at scale — with no infrastructure to build or maintain. Sessions go from creation to a first response in under a couple of seconds, and resume paused work in about 300 milliseconds.
Run an agent locally and the session dies when the laptop sleeps or the process stops. There’s no way to run many sessions in parallel, no state a teammate can pick up, and no clean way to pause a long task and resume it later.
Move to a cloud VM and the agent survives longer, but now someone has to build and maintain everything around it: session persistence, credential handling, tool integrations, and handoff between devices or teammates. Every team ends up building a slightly different version of the same plumbing.
Production agents need to touch real systems: repositories, tickets, databases, SaaS tools. Connecting each one independently means a new credential, permission model, and retry system for every integration, with no consistent way to trace what an agent actually did or what it cost.
DigitalOcean Managed Agents is made up of two components, vertically integrated, that you can start using today.
Durable execution layer
The execution layer that runs your agent sessions in the cloud and keeps them running, without you provisioning or maintaining infrastructure.
Supports coding harnesses (Claude Code, Codex CLI, OpenCode), general-purpose agents (Hermes), and agents built with frameworks like LangGraph.
Governed tool access
The tool-access layer that gives agents governed access to 16,000+ tools through a single managed MCP endpoint.
DigitalOcean Managed Agents is harness- and framework-agnostic. Bring the agent you’re already running.
No rebuilding around a DigitalOcean-specific harness. No lock-in to one model provider.
No more stitching together a harness, sandbox, model, or search API. One price and one bill, covers all of it.
Developers curious about agents
For teams running agents regularly
For production fleets
*Available to new DigitalOcean customers only. Existing customers can start on any paid plan.
Sessions persist across devices and teammates, survive a closed laptop, and can run many at once instead of one at a time.
Bring any supported harness, any model, or your own OCI image. No lock-in to a single vendor’s runtime.
Credentials are brokered at execution time and never reach the model. Every tool call is traced, permissioned, and attributed to cost.
CPU billing follows actual consumption per second, so an idle agent’s charge falls to zero instead of paying for allocated capacity.
Run Managed Agents alongside the rest of DigitalOcean's platform, no stitched-together vendors, fragmented billing, or separate infrastructure to manage.
DigitalOcean Managed Agents is DigitalOcean’s platform for running AI agents in production. It’s made up of DigitalOcean Harness Runtime, the durable execution layer that runs and persists agent sessions, and Action Gateway, the tool-access layer that gives agents governed access to the systems they need to complete work. Together, they let you run agents durably and give them the tools to finish real tasks, without building the surrounding infrastructure yourself.
DigitalOcean Harness Runtime runs specialized coding agents (Claude Code, Codex CLI, OpenCode), general-purpose agents (Hermes), and agents built with frameworks like LangGraph and CrewAI. You can also bring your own agent as a standard OCI container image and run it as a reusable template.
Managed Agents is available in Public Preview to all customers who opt in to the preview terms. You can create sessions using Claude Code, Codex CLI, OpenCode, Hermes, or LangGraph through the web console or the CLI (doctl), and configure tool access through Action Gateway using the DigitalOcean SDK, Inference Engine, or any MCP-compatible client.
The session keeps running in the cloud. It finishes the work it was given, then pauses and snapshots its state. You can reattach later, from the same device or a different one, and pick up exactly where the agent left off — resume takes about 200 milliseconds.
Yes. You can checkpoint a session at a known-good point, then fork a new, independent session from it. This lets you try a few different approaches in parallel and keep the one that works, without losing the original.
Credentials are brokered at execution time and never reach the model, the agent prompt, or the sandbox. You can connect credentials as a static API key, a shared OAuth app for the whole agent, or per-user OAuth, where each end user authorizes their own account.