Best VPS Providers for Hosting n8n Agents in 2026 (Compared)

n8n Cloud is convenient, but self-hosting on a VPS is cheaper at scale, keeps credentials on your own infrastructure, and removes execution limits — which matters once AI agents run long, looping workflows. Here’s what to buy and where.

What specs do n8n agents actually need?

• Minimum for light use: 2 vCPU / 4 GB RAM. Fine for scheduled syncs and a low-traffic chatbot.
• Sweet spot for AI agents: 4 vCPU / 8 GB RAM. Agentic loops, Whisper transcription, and local vector stores are memory-hungry; 4 GB swaps under load.
• Running local LLMs (Ollama) too? You want 16–32 GB RAM and ideally a GPU box — at that point compare against a dedicated server or just calling a hosted model API.
• Storage: 40–80 GB NVMe is plenty; workflow JSON is tiny, execution history is what grows. Set EXECUTIONS_DATA_PRUNE=true so old runs don’t fill the disk.

The providers, compared

Hetzner — best value overall

CX22/CX32 class instances (2–4 vCPU, 4–8 GB RAM) at prices competitors can’t touch, on fast NVMe in EU and US regions. The default recommendation for self-hosted n8n in 2026. Minor downside: account verification can be slow for first-time customers.

DigitalOcean — easiest setup

Slightly pricier per GB, but the 1-Click n8n droplet, predictable billing, and excellent docs make it the fastest path from zero to running. Great if you’d rather pay a small premium than SSH through setup.

Contabo — cheapest raw specs

Very low prices for high RAM allocations, which tempts agent builders. Trade-off: older hardware generations and support that tests your patience. Acceptable for staging; think twice for production support bots.

Hostinger — beginner-friendly with 1-click n8n

Hostinger’s KVM VPS plans (NVMe storage, AMD EPYC on higher tiers, data centers across the US, EU, and Asia) pair low entry prices with an hPanel-managed experience and a 1-click n8n app template, so you can spin up a working instance without touching Docker Compose manually. Weekly snapshots/backups are built in. Resource caps are stricter than Hetzner’s at the same price, so size one tier up if your agents run heavy RAG or transcription loops.

OVHcloud — solid EU alternative

Competitive pricing with strong European data-center coverage for GDPR-sensitive deployments. Control panel is less polished than Hetzner or DO, but the network is reliable.

Vultr / Linode (Akamai) — dependable middle ground

Hourly billing, wide region choice, and one-click Docker images. Pick whichever has a data center closest to your users — latency matters for chat agents.

Docker setup checklist (any provider)

1. Provision Ubuntu 24.04 LTS, create a non-root sudo user, enable UFW (allow 22, 80, 443 only).
2. Install Docker + Compose plugin; deploy n8n with a Compose file pinning a specific version tag — never latest in production.
3. Put Caddy or Traefik in front for automatic HTTPS; n8n webhooks and the chat widget require TLS.
4. Mount a persistent volume for /home/node/.n8n, and back it up nightly (workflow JSON + credentials live there).
5. Set N8N_ENCRYPTION_KEY explicitly and store it off-server — lose it and your credentials are unrecoverable.
6. Enable pruning (EXECUTIONS_DATA_PRUNE, EXECUTIONS_DATA_MAX_AGE=168) and watch disk with a simple cron alert.

The total cost picture

A 4 vCPU / 8 GB VPS runs roughly $6–24/month depending on provider — against per-execution pricing that climbs fast once agents loop. Add your LLM API spend on top (usually the bigger line item), and self-hosting wins for anything beyond experimentation.

What to run on it

Once the server is up, you don’t need to design agents from scratch. Import proven workflows from the free n8n agents templates library — RAG chatbots, triage agents, enrichment flows — reconnect your model and service credentials, and you’re live on your own VPS the same day.