DigitalOcean Managed Agents is now in public preview. Run Claude Code, Codex, or your own LangGraph agent in a runtime environment that pauses when idle. Put its tools behind one governed endpoint, and pick from 75+ open and proprietary models. One cloud, one bill. Prompts to get started available... in the blog:show more

Harness Runtime gives the agent a super fast machine. One doctl command puts Claude Code, Codex, LangGraph, or more into an isolated environment.

Sessions are durable and fast to spin up or down. Pause one on Friday, resume it Monday with the workspace and the whole conversation still there. You don't pay for it while it sits, and it resumes in 316ms.

Checkpoints snapshot the machine and the agent's memory together. Fork one session three ways, throw three approaches at the same refactor, keep the one that works.

Action Gateway gives it tools. @ExaAILabs, @Supabase, and @Box are all providers behind one MCP endpoint. The credentials resolve when a call runs, exist for that one call, and never reach the agent or the sandbox.

@supabase @Box One more doctl command gives you direct access to 75+ open and proprietary models, with advanced features like intelligent model routing and synthesizing the best results of multiple models.

.@qencode built a support-triage agent on Managed Agents. Their agent reads each incoming request, weighs urgency, sentiment, and account value, then opens or updates a Jira ticket, and kicks anything it isn't sure about to a human. Qencode CEO @muradmordukhay reports that the new agent took 1 day to complete and is saving up to a full day per week.

“Pauses when idle” gotta test it out

Let us know when you think!

Love this. Pause Friday, resume Monday, same conversation. Finally treating agents like work, not like a process I have to babysit.

$DOCN remains one of my favorite ways to play agentic AI as the company continues to build layer that routes each workload to best model for price performance. Long-term AI winner.

the pause-when-idle part might be the first cloud feature my side project budget actually understands

Love to hear it! Your budget shouldn't constrain what you're able to build.

Let's get building!!!!

forking one session three ways and keeping the one that works is actually really useful

When you fork a session 3 ways, do all 3 share the same tool credentials or does each fork get its own?

What does the cost per completed agent task look like versus stitching together separate sandbox, inference, storage, and tool providers?

finally, a runtime that doesn't burn compute while idling. looks useful.

Keep us posted on what you build!

Yeah, I think it is the future to some extent. There's no reason for everything to run local. Everything's gonna be more and more cloud-based going forward, it seems.

👀

Nice move, DigitalOcean! The idle pause actually makes a lot of sense for agents that don’t need compute running all the time.

Per-call credentials that never touch the agent or sandbox is the only sane MCP default. Long-lived tokens in the prompt are a bug wearing a feature badge.

The pause-when-idle runtime is the part I'd test first: on resume does it keep context and open tool sessions, or cold start and re-read state? That's where long agent runs quietly get expensive.

You're lucky there isn't a good option for our location to migrate away. Your dashboard is fucking terrible now with AI shoved down our throats everywhere. Glad to see even more... yay

Pauses when idle 🙏

yo this is sick, how did you manage long sessions with claude code?

Going to use this to build an ai version of @BacardiCapital