DigitalOcean engineering blog

DigitalOcean introduces MicroVMs, lightweight Firecracker-based virtual machines for isolated, bursty workloads such as AI agent sessions, code sandboxes, and short-lived jobs. The private-preview service automatically pauses idle VMs and resumes them with memory, files, and processes intact, while checkpoints reduce startup overhead.

DigitalOcean introduces Agent Droplets, a bundled pricing plan for Managed Agents that combines dedicated microVM compute, hosted inference, persistent storage, and governed access to more than 16,000 tools. The post details plan tiers, usage estimates, security features, automatic pausing, and prepaid billing behavior for agent workloads.

DigitalOcean explains how its Managed Agents architecture keeps credentials outside agent contexts through execution-time brokering, scoped secrets, gateway policies, and isolated microVMs. The post details how these controls limit exfiltration, constrain agent swarms, and provide auditable, policy-driven access to connected systems.

DigitalOcean introduces Managed Agents, an integrated runtime combining isolated Firecracker microVM sandboxes, model inference, persistent data, governed tool access, and observability. The post explains its burst-based pricing, credential brokering, durable sessions, and architecture for deploying secure, stateful AI agents.

DigitalOcean introduces Managed Agents, combining Firecracker microVM-based durable sessions with governed access to more than 16,000 tools through a managed MCP endpoint. The post explains pause/resume and forking, active-CPU billing, security controls, lifecycle APIs, and benchmarked startup and resume performance.

DigitalOcean announces general availability of Managed Databases Advanced Edition for MySQL and PostgreSQL, targeting high-concurrency and mission-critical workloads. The release adds rapid failover, faster storage scaling, deeper observability, VPC connectivity, and measured improvements in throughput, latency, and storage cost.

DigitalOcean argues that open models, agents, harnesses, data technologies, and infrastructure can prevent AI capabilities from becoming concentrated among a few providers. The post promotes its Open Intelligence vision and announces an upcoming summit focused on building, serving, and operating an open AI stack.

DigitalOcean is becoming the Omacom Foundation’s founding corporate patron and the compute provider for Omarchy, a keyboard-first Linux desktop. Omarchy’s bursty pipeline for packaging, agent-based pull-request reviews, and QA testing will provision and destroy Droplets on demand.

DigitalOcean introduces v5 Droplets powered by 5th Gen AMD EPYC processors, delivering up to 30% higher per-core performance for demanding workloads. The new generation supports independent vCPU, memory, and storage selection, dedicated or shared CPU options, and deployment through the console, API, or Kubernetes node pools.

DigitalOcean introduces M.A.R.S., a managed runtime for persistent, scalable AI-agent sessions and multi-tool workflows. It combines isolated Firecracker microVM execution with centrally governed access to services, OAuth credentials, approvals, and audit controls, while supporting multiple agent harnesses and frameworks.

DigitalOcean details how it patched its hypervisor fleet against the Januscape KVM escape vulnerability and the AMD Safe RET issue without confirmed customer impact. The post explains livepatching, workload evacuation, capacity reclamation, staged rollouts, detection, and the operational practices that enabled repeatable fleet-scale remediation.

DigitalOcean explains how cache-aware model routing can reduce inference cost and latency for long-running agent sessions. The post details the X-Model-Affinity header, routing-budget policy, automatic affinity detection, and new cache-efficiency analysis tools, including the trade-offs of switching models.

DigitalOcean details how it deployed and optimized the 2.78-trillion-parameter Kimi K3 model across NVIDIA and AMD GPU servers. The post covers memory planning, heterogeneous inference, vLLM tuning, vendor verification, dynamic tools, reasoning controls, streaming compliance, and proxy-layer fixes that improved correctness and performance.

DigitalOcean introduces model synthesis for its Inference Engine, orchestrating parallel model panels and a synthesizer through one API call. Benchmarks show an open-source GLM 5.2 and Kimi K2.6 configuration outperforming tested single models at substantially lower cost, with practical presets and direct configuration examples.

DigitalOcean announces GPU pricing changes taking effect August 1, 2026, covering on-demand and 12-month reserved GPU Droplets. Existing reserved contracts retain their current rates, while active workloads and renewed plans will be billed at updated prices.

DigitalOcean’s Managed Weaviate public preview provides production-ready vector databases with automated backups, upgrades, high availability, autoscaling, monitoring, and TLS. The service preserves Weaviate API compatibility, uses predictable pricing from $20 per month, and enables RQ8 compression for lower memory usage in RAG, semantic search, and agentic applications.

Engineering leaders from Workato, Hippocratic AI, and ISMG share production lessons for scaling AI inference, including managing P99 latency, controlling agent permissions, and preparing policy-aware infrastructure. The panel emphasizes that reliable, cost-efficient AI at scale is primarily an architecture and governance challenge.

DigitalOcean Evaluations provides structured LLM-as-a-Judge testing for models, fine-tuned deployments, BYOM imports, and inference routers using teams’ own datasets and rubrics. It tracks quality, latency, token usage, and cost, with versioned datasets, reusable presets, API/SDK and MCP triggers for repeatable production and CI workflows.

DigitalOcean’s Codex plugin provisions persistent, Codex-ready Droplets directly from the Codex app through OAuth and natural-language prompts. The post explains the provisioning and SSH connection flow, Marketplace setup, remote handoff capabilities, and mobile monitoring options.

DigitalOcean introduces Server-Side Tools in public preview for its Inference Engine, enabling models to use web search, URL fetching, Knowledge Bases, MCP servers, and Anthropic or OpenAI tool conventions within inference requests. The post explains access, Web Mode, supported workflows, and how the feature reduces the need for separately managed tool infrastructure.