A server for n8n and automations
n8n is an automation platform you can install on your own server: workflows connect a CRM, email, spreadsheets, bots and AI model APIs. n8n itself is undemanding, but without a couple of settings the server eventually runs out of memory or disk. Here is which VPS to take and what to configure right away.
The short answer. A KVM VPS with 2 vCPU and 4 GB of memory is enough for n8n: n8n in Docker, PostgreSQL alongside, a domain with HTTPS. Memory matters more than cores: large data in workflows is what consumes it. Set up execution history pruning straight away and keep the encryption key together with the database backup.
What uses resources
Most of the time a workflow waits for an external API to respond and the CPU is idle, so shared cores are usually enough (how they differ from dedicated ones is covered in A host's plan grid: vCPU, fair use and burst). Memory, according to the n8n documentation, grows with the amount of JSON and binary data in a workflow, the number of nodes, the Code node, manual executions (n8n copies data for the editor) and the number of workflows running at once. If workflows process exports of thousands of rows or files, take 4–8 GB; Node.js gets more memory with the --max-old-space-size option.
Database and execution history
By default n8n keeps workflows, credentials and history in SQLite. For a production server PostgreSQL is more convenient: as of July 2026, versions 16, 17 and 18 are supported. Execution history stores the data of every run, and n8n warns that the database can fill the whole disk. By default, executions older than 14 days (EXECUTIONS_DATA_MAX_AGE=336 hours) and beyond 10,000 are deleted; with hundreds of executions an hour, lower these values. More on databases on a VPS in A database on a VPS: PostgreSQL or MySQL.
Webhooks and HTTPS
Services that trigger workflows by webhook call the server over HTTPS, so you need a domain, a certificate and a reverse proxy (nginx, Caddy or Traefik). n8n listens on port 5678 and the proxy on 443, so the external webhook URL must be set in the configuration: otherwise n8n will show and register the wrong address with external services. Webhooks need a dedicated IPv4: not every service will accept a server without one, see IPv6-only VPS: savings and limitations. How to run containers is covered in Docker on a VPS: getting started; the selection: VPS for Docker.
Encryption key and backups
Passwords and tokens in n8n are encrypted with a key that is generated on first launch and stored in the ~/.n8n folder (in Docker, in the container volume). Without it a database backup is useless: the credentials cannot be decrypted. Set the key yourself via N8N_ENCRYPTION_KEY and keep it away from the server, and back up the database daily — see VPS backups: strategies and tools.
When one server is not enough
For heavy loads n8n has a queue mode: the main process accepts executions and separate workers run them, up to 10 jobs each by default. It needs Redis and PostgreSQL — the mode does not work with SQLite — and the encryption key must be the same on all workers.
Where to put it and what it costs
Put the server close to the services your workflows talk to: for foreign APIs and Telegram bots, Europe (for the reasons, see VPS for a Telegram bot: what the bot needs and how to pick a plan); for 1C, Bitrix24 and VKontakte, Russia. In the accsly.io catalog on 17 November 2026 there are 2,576 plans with exactly 2 vCPU and 4 GB; the monthly median with monthly billing is €16.31 for plans priced in euros, $24.99 in dollars and 1,080 ₽ in roubles. Examples as recorded: Hetzner CPX22 in Germany and Finland — €19.99, Vultr — $20, DigitalOcean Basic — $24, Timeweb Cloud-50 in Russia — 3,840 ₽. The selection: VPS for n8n. If an AI agent that edits and tests code by itself will run alongside, it needs more memory — see VPS for Claude Code.