Quick answer: For one always-on AI agent or an n8n instance, 2 to 4 GB of RAM is usually enough. Add about 1.2× the model size in RAM if you also run a local model with Ollama, and use a GPU cloud for models of 14B parameters or more. Unmanaged clouds (Hetzner, DigitalOcean, Vultr) are cheapest per month with no renewal jump. Hostinger VPS is the easiest first VPS.
PillarBest VPS for AI agentsWhere to run OpenClaw, n8n and Ollama 24/7How-toHost an AI agent on a VPSStep-by-step OpenClaw setupGuideSelf-hosted AI agent stackn8n + Ollama + Open WebUI + Postgres on one VPSGuiden8n VPS requirementsRAM and CPU for self-hosted n8nGuideOllama VPS requirementsCPU vs GPU, RAM per model sizeGuideVPS for Discord/Telegram botsCheap always-on optionsGuideCheap VPS for beginners7 plans under $12ComparisonHetzner vs ContaboPrice, RAM and reliabilityReviewHostinger VPS reviewPrices, renewals, AI templatesGuideVPS vs shared hostingWhich one you actually needToolVPS / AI-agent sizerRAM, vCPU and disk for your workloadToolLatency testWhich region is fastest from your city
Frequently asked questions
How much RAM does an AI agent VPS need?
An agent that calls a hosted LLM API needs about 2 GB of RAM. With n8n and Postgres, plan for 4 GB. Running a local 7–8B model with Ollama adds roughly 6–8 GB. Our VPS sizer estimates your exact mix.
Do I need a GPU VPS?
Not for agents that call APIs. For local models of 14B parameters or more, a CPU-only VPS is very slow. Rent an hourly GPU instead.
Prices are published list prices from official provider pages, last checked 21 Oct 2026. They are not GetBestHost lab benchmarks. Some links are affiliate links (marked "sponsored"); they never change our rankings. See the affiliate disclosure and methodology.