Budget automation guide

Cheap VPS for AI Agents & Automation

Most "AI agents" are orchestration plus API calls. That runs on a small server. Here is how to size it without wasting money.

Affiliate disclosure: some links below are affiliate links. We may earn a commission at no extra cost to you.

What an AI agent actually runs

An agent is usually a loop: receive a task, call a model API, choose a tool, run it, and repeat until done. The heavy computation happens on the model provider's side. Your server handles orchestration, queues, HTTP calls, and storage.

That means a 1–2 GB VPS is enough for many agents. The exceptions are browser automation, local model inference, and heavy data processing, which need more RAM and CPU.

Self-hosting n8n and similar tools

n8n, Flowise, and similar automation tools run comfortably on a small VPS for light workloads. The common mistake is letting execution data grow without limits, which bloats the database and eventually the disk.

Set retention for executions, use Postgres instead of SQLite as you scale, and back up the database. A 2 GB VPS with a small Postgres instance handles a surprising amount of automation.

Browser automation is the memory hog

Playwright or Puppeteer with Chromium can use several hundred MB per browser context. Running multiple pages in parallel quickly pushes memory past 2 GB. If your agent browses websites, budget 2–4 GB and limit concurrency.

Headless browsers also consume CPU and disk for caches. Use a single browser instance with multiple contexts where possible, and recycle contexts to avoid leaks.

Scraping and IP reputation

Cheap VPS providers use datacenter IP ranges, which many sites block or rate-limit. A faster or cheaper server will not fix that. Options include slower request rates, rotating proxies, or using an official API instead of scraping.

Proxies usually cost money and are a third-party dependency. If your budget is tight, prefer APIs and public datasets, and respect robots.txt and terms of service.

How much server do you need?

WorkloadRAMvCPUNotes
API-based agent1–2 GB1Orchestration and queues
n8n / automation1–2 GB1–2Postgres as workflows grow
Browser automation2–4 GB2Limit Chromium contexts
Scraper1–2 GB1–2IP blocks are the real limit
Local 7B model8–16 GB4+GPU is much faster

Budget-friendly options

For a paid server, Hetzner remains the best value for CPU workloads. Contabo offers high RAM for a low price but with slower I/O. Oracle Cloud's Always Free tier can work for prototypes, subject to capacity and verification.

Hetzner

Best price-to-performance for agents and automation. EU regions.

Check Hetzner

Contabo

High RAM plans at low prices. Check I/O performance and setup fees.

Check Contabo

Oracle Cloud Free

Always Free tier for prototypes. Capacity is not guaranteed.

Check Oracle Free

Vultr

Many regions, predictable hourly billing, easy to resize.

Check Vultr

Cost control

  • Start with 1–2 GB and resize only when you see pressure.
  • Set execution data retention for automation tools.
  • Cap browser concurrency and recycle contexts.
  • Monitor disk usage; logs and caches grow quietly.
  • Stop test instances instead of leaving them running.

FAQ

How much RAM does an AI agent need?

If the agent calls external APIs, 1–2 GB is usually enough. Browser automation with Playwright needs 2–4 GB. Running a local model is a separate, much larger requirement.

Can I run n8n on a cheap VPS?

Yes. n8n runs on a 1 GB VPS for light workflows. Use 2 GB or more once you have many workflows, and move from SQLite to Postgres as usage grows. Set execution data retention so the database does not grow forever.

Why do scrapers get blocked on cheap VPS?

Many sites block datacenter IP ranges. A cheaper VPS does not fix that. You may need rotating proxies, slower request rates, or a different data source. Proxy access usually costs extra.

Is there a free VPS for AI agents?

Some providers offer free tiers, such as Oracle Cloud Always Free, subject to capacity and verification. Free tiers are useful for prototypes, but availability and performance are not guaranteed, so do not depend on them for production.

Last updated: November 2026. Free tiers and pricing change frequently; confirm details with each provider.