DigitalOcean Agent Droplets Bundle AI Agents Into One Subscription

BROOMFIELD, Colo. — DigitalOcean announced on October 1 a new way for companies to pay for AI agent infrastructure: Agent Droplets, a single monthly subscription that bundles compute, memory, storage, inference, and tool access into one predictable bill. The move packages the company's Managed Agents platform — which entered public preview on September 22 — into two subscription tiers aimed at teams moving from agent experiments to production.

Agent Droplets come in two sizes: Pro at $50 a month and Team at $200 a month, with discounts of 15% and 20% respectively across all included resources. For developers, the pitch is simplicity. Instead of assembling a runtime, a tool gateway, inference, and storage from separate services and watching the meter run on each, teams pick a size and start building.

One subscription for the whole agent stack

Under the hood, an Agent Droplet bundles the capabilities DigitalOcean introduced with Managed Agents. The first is the Harness Runtime, a lightweight microVM that gives each agent a persistent, isolated compute environment. Sessions can be paused, resumed, or forked, and idle agents pause automatically — a design choice aimed at killing the idle-compute bills that plague always-on agent setups.

The runtime supports popular coding agents such as Claude Code, Codex CLI, and OpenCode, alongside general-purpose agents like Hermes and custom agents built with LangGraph. DigitalOcean says agents can connect securely to internal company services from inside the hardened environment, and conversational history and working state persist across sessions.

The second piece is the Action Gateway, the agent's interface to the outside world through a unified MCP endpoint. The gateway exposes more than 16,000 tools — web search, browser automation, DigitalOcean's own infrastructure APIs, and connectors for GitHub, HubSpot, and Stripe. Access is governed centrally: administrators manage permissions, sensitive operations require human approval, and the gateway handles rate limiting, retries, backoff, and timeouts for multi-agent workloads.

Rounding out the bundle are serverless inference and persistent agent memory and storage. Developers can create sessions through DigitalOcean's Cloud Console or the doctl command-line tool, and configure tool access with the company's SDK or any MCP-compatible client. New users receive a $5 credit to launch their first agent session, according to the company's launch announcement carried by Business Wire.

Predictable pricing for unpredictable workloads

Billing is where DigitalOcean is trying to differentiate. Alongside the new subscription tiers, the company introduced an Inference and Agents balance: customers can prepay $5 to $500 at a time or enable auto-reload, with funds that never expire. Paying through a monthly plan carries discounts on all included resources, in contrast to the metered pricing that dominates agent platforms.

Industry observers have pointed to cost as the biggest friction in agent adoption. Former Meta AI engineer Elvis Saravia praised DigitalOcean's focus on performance and cost reduction, noting that many competing agent management platforms are cost-prohibitive, in coverage by TechGig. Ryan Martin, a platform architect at Seaotter, described the combination of pause-when-idle runtimes and governed tooling as the operations pattern that actually scales.

Chief Product and Technology Officer Vinay Kumar framed the launch as a repeat of the company's original playbook, in the Agent Droplets announcement. Fourteen years ago, he said, the humble Droplet made the cloud something a single developer could understand and afford — and a generation of companies got built on it. The Agent Droplet, in his telling, is the same idea applied to agents: one subscription for everything an agent needs, so a team with an idea at 11 p.m. can have it running before they go to sleep.

Why agent infrastructure is having a moment

DigitalOcean is not alone in betting that agents need their own infrastructure layer. Docker's Cloud Sandboxes offer similar microVM isolation and persistence, though DigitalOcean's pitch leans harder toward production orchestration and tool support. The launch also lands days after OpenAI's DevDay, where always-on agent products took center stage — a sign that the industry's center of gravity is shifting from chatbots toward persistent, background agents that need somewhere to live.

Early adopters are already wiring the platform into real products. Robert Brennan, co-founder and CEO of OpenHands, said the runtime and gateway give his team a foundation to scale while focusing engineering effort on automating complex workflows. Spenser Skates, co-founder and CEO of Amplitude, said the company will use the Harness Runtime and Action Gateway to power its agentic product team, citing fast startup and scaling, intelligent tool discovery, and CPU pricing tied to consumption.

The broader trend is consolidation across the agent stack. New observability products like Mixpanel Agent Intelligence aim to measure what agents do in production, while shared-workspace tools like Cua Spaces give teams a common desktop for agent work. DigitalOcean's bet is that the next consolidation happens one layer down: the runtime itself. If agents are going to run around the clock, someone has to host the machines, govern the tool calls, and make the bill make sense.

For now, Managed Agents and Agent Droplets are in public preview, and the subscription pricing is a preview offer. But the direction is clear: the agent economy is getting its own infrastructure stack, and the cloud providers that make it boring — predictable, governed, affordable — may end up owning it.