DigitalOcean Managed Agents
by DigitalOcean
Managed agent runtime with microVM sandboxes, 16,000+ governed tools and serverless inference, billed on active CPU
DigitalOcean Managed Agents is a managed cloud platform for running AI agents that bundles isolated microVM runtimes, a governed tool gateway and serverless model inference. It is for developers and AI-native product teams who want to run Claude Code, Codex CLI, LangGraph or custom agents in production without assembling sandboxes, tool auth and inference themselves.
DigitalOcean Managed Agents entered public preview on September 22, 2026. It is built from two services wired into DigitalOcean's Inference Engine. Harness Runtime runs each agent in a hardware-isolated Firecracker microVM with built-in Chromium and a coding sandbox, exec APIs and port forwarding, and lifecycle APIs to pause, resume and fork sessions; DigitalOcean reports resume-to-ready in 305 milliseconds and session creation to first response in 3.3 seconds. It accepts coding harnesses such as Claude Code, Codex CLI and OpenCode, general agents such as Hermes, LangGraph and LangChain agents, and custom OCI containers. Action Gateway is a managed MCP endpoint exposing more than 16,000 tools from over 500 providers, including GitHub, Jira, Stripe, Snowflake, PagerDuty, Box and Supabase, plus web search, web fetch and browser automation; credentials are brokered at execution time and never reach the model or sandbox, and sensitive actions can require approval. The Inference Engine routes across more than 75 open and proprietary models. Runtime billing counts only active CPU at $0.044 per vCPU-hour and memory at $0.0095 per GB-hour, so waiting on a model or tool costs no CPU. On October 1, 2026 DigitalOcean added Agent Droplets, monthly bundles at $50 (Pro) and $200 (Team) with unlimited agents and seats. Early users named by DigitalOcean are OpenHands, Qencode and Amplitude. It competes with AWS Bedrock AgentCore, Microsoft Foundry, Google Cloud and Cloudflare's agent platform.
A platform or AI engineering lead at a startup or mid-size software company that wants to run coding and support agents in isolated sandboxes with predictable cloud spend.
One provider supplies the sandbox, tool credentials and model inference for an agent, billed on active CPU or a flat monthly bundle instead of several metered vendors.
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Usage-based, Subscription
- Target Market
- CTOs, Platform Engineers, AI Engineers, Enterprise Developers
- Deployment
- Cloud-only, API-based
- Founded
- 2011
- Headquarters
- Broomfield, Colorado, United States
- Team Size
- 500+
Key Features
- ✓Harness Runtime
Firecracker microVM sandboxes with Chromium and a code sandbox that can pause, resume and fork sessions, keeping agent state without paying for idle compute.
- ✓Action Gateway
A managed MCP endpoint to 16,000+ tools from 500+ providers with credentials brokered at execution time, so secrets never reach the model or sandbox.
- ✓Approval controls
Centralised permissions and approval gates for sensitive tool operations give security teams a checkpoint before an agent acts.
- ✓Inference Engine
Serverless access to 75+ open and proprietary models with routing across them for cost and latency, from the same cloud as the runtime.
- ✓Active-CPU billing
Charges $0.044 per vCPU-hour only while CPU is in use, so time spent waiting on model responses or tool calls does not accrue CPU cost.
- ✓Agent Droplets
Monthly bundles at $50 and $200 with 15% or 20% off agent usage and unlimited agents and seats, for predictable budgeting.
- ✓Built-in access controls
SSO, multi-factor authentication, role-based access, audit logs, cloud firewalls and DDoS protection are included on all Agent Droplet tiers.
Capabilities
Use Cases
- •Hosted coding agents
Engineering teams run Claude Code or Codex CLI sessions in isolated microVMs, forking sessions for parallel subagents and pausing them between tasks.
- •Support ticket triage
Qencode runs a support-triage agent that routes tickets automatically, which its CEO says saves four to eight hours a week.
- •Agents embedded in SaaS products
Product companies such as Amplitude use the runtime and tool discovery to ship customer-facing agents without building their own sandbox layer.
- •Governed operations agents
Ops teams let agents act in GitHub, Jira or PagerDuty through the gateway, with approval required before sensitive changes go through.
Ideal For
Best For
- ✓Running Claude Code, Codex CLI or OpenCode agents in isolated cloud sandboxes
- ✓Long-lived agents that wait on models or tools much of the time and benefit from CPU-only-when-active billing
- ✓Teams that need agents to call SaaS tools without exposing API credentials to the model
- ✓Small teams wanting a flat monthly agent budget through Agent Droplet plans with unlimited seats
Not Ideal For
- ✗Production workloads that need an SLA today, since Managed Agents and Agent Droplets are in public preview without guaranteed production-level performance
- ✗Teams standardised on frontier models for every call, because Claude and GPT usage is billed at list price outside the Agent Droplet discounts
- ✗Enterprises that need HIPAA coverage or service-specific compliance attestations confirmed before a pilot, which the launch materials do not state
Deployment
Market Analysis
Pros
- ✓Idle wait time is free on CPU, which suits agents that spend much of a session waiting on models
- ✓Credentials are brokered at execution time and never exposed to the model or sandbox
- ✓Framework-agnostic: runs Claude Code, Codex CLI, LangGraph, CrewAI and custom OCI containers
- ✓Early adopters include OpenHands and Amplitude, both named with on-record quotes
Cons
- ✗Public preview only, with production-level performance explicitly not guaranteed
- ✗Frontier models are excluded from bundle discounts, so Forkast notes the intelligence stays a premium variable cost
- ✗Little independent practitioner feedback yet: the Hacker News launch post drew 4 points and no comments
- ✗Speed and accuracy figures (305 ms resume, 99.3% tool matching, 37% lower TCO) are vendor-reported benchmarks
Pricing
Pay as you go
$0.044/vCPU-hour + $0.0095/GB-hour
- ✓Active-CPU billing
- ✓Snapshots $0.005/GiB-month
- ✓Inference and tools metered separately
Agent Droplet Pro
$50/mo
- ✓15% off agent usage
- ✓Unlimited agents and seats
- ✓SSO, RBAC, audit logs
Agent Droplet Team
$200/mo
- ✓20% off agent usage
- ✓Unlimited agents and seats
- ✓SSO, RBAC, audit logs
Runtime is metered on active CPU at $0.044 per vCPU-hour and memory at $0.0095 per GB-hour, with snapshots at $0.005 per GiB-month; inference, storage and third-party tools are billed separately. Agent Droplet bundles cost $50 or $200 a month, new users get $5 of credit, and frontier models such as Claude and GPT stay at list price outside the bundle discounts.
Security & Compliance
Sources
This page was written from 9 sources, 6 on domains other than digitalocean.com.
- 1.digitalocean.com — managed agents public previewvendor
- 2.digitalocean.com — introducing agent dropletsvendor
- 3.investors.digitalocean.com — default
- 4.digitalocean.com — certification reportsvendor
- 5.forkast.news — digitalocean turns agent infrastructure into a commodity sku
- 6.forkast.news — digitalocean managed agents the agent as the new front door
- 7.hostingjournalist.com — digitalocean launches agent droplets with monthly plans
- 8.en.wikipedia.org — DigitalOcean
- 9.hn.algolia.com — search
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Modular
MAX inference framework and Mojo language for serving AI models on NVIDIA, AMD and other chips
ZML/LLMD
Free, Python-free LLM inference server that runs open models on NVIDIA, AMD, Google TPU, Intel and Apple chips from one binary
Lambda
GPU cloud and AI factories for training and inference — on-demand NVIDIA instances to single-tenant superclusters
Anyscale
Managed Ray platform for scaling AI data processing, training, inference and RL across thousands of GPUs on any cloud