Run n8n, Ollama or your own chatbot on a Hostinger VPS: which KVM plan fits which model size, setup steps, security tips and when to use an API instead.
> **Disclosure:** some links on this page are affiliate links. If you buy or book through them we may earn a commission, at no extra cost to you. It never changes what we recommend.
**Short answer:** A Hostinger VPS is a cheap, sensible home for n8n automations and small [AI chatbots](/blog/meta-s-muse-ai-vs-chatgpt-grok-claude-which-chatbot-wins-in-2026) that call an API, and KVM 2 (2 vCPU, 8 GB RAM) handles that comfortably. For running your own model with Ollama, pick KVM 4 (16 GB RAM) or above: an 8B model fits, but these are CPU-only servers, so replies will be slow. If you need fast answers from a strong model, an API will usually beat a self-hosted model on both quality and cost.
This guide covers which plan suits which AI job, how to set each one up, and when to skip self-hosting entirely. Prices are at the time of writing (October 2026); they change often and depend on the term you prepay.
## Hostinger's VPS plans in October 2026
[Hostinger's VPS page](https://www.hostinger.com/vps-hosting) lists four KVM plans. All include:
- AMD EPYC processors and NVMe storage;
- free weekly backups;
- a firewall and a 1 Gbps network;
- an AI web terminal;
- a free domain for one year.
| Plan | vCPU | RAM | NVMe | Bandwidth | Intro (US) | Renews (US) | Intro (India) | Renews (India) |
|---|---|---|---|---|---|---|---|---|
| KVM 1 | 1 | 4 GB | 50 GB | 4 TB | $6.49/mo | $11.99/mo | ₹599/mo | ₹999/mo |
| KVM 2 | 2 | 8 GB | 100 GB | 8 TB | $8.99/mo | $14.99/mo | ₹799/mo | ₹1,199/mo |
| KVM 4 | 4 | 16 GB | 200 GB | 16 TB | $12.99/mo | $28.99/mo | ₹1,099/mo | ₹2,399/mo |
| KVM 8 | 8 | 32 GB | 400 GB | 32 TB | $25.99/mo | $49.99/mo | ₹2,199/mo | ₹4,399/mo |
Indian prices are from [Hostinger India's VPS page](https://www.hostinger.com/in/vps-hosting). UK prices start at £4.99 for KVM 1. Hostinger says renewal rates apply for two years and that plans are paid upfront. [Website Builder Expert's 2026 pricing breakdown](https://www.websitebuilderexpert.com/hosting-websites/hostinger-pricing/) lists the intro VPS prices against a 24-month term.
**Note what is missing: there is no GPU.** That is fine for automations and API-based bots, but it limits local models (more below).
## One-click templates for AI builders
You do not have to start from a bare Linux server. Hostinger's VPS page lists one-click templates including:
- **n8n**, Docker, Claude Code and Codex CLI;
- **Supabase**, Redis, Grafana, Uptime Kuma, WireGuard and Nextcloud;
- **app platforms** such as Coolify and Dokploy;
- **an LLM template** and a catalogue of AI agents.
Hostinger's own [Ollama tutorial](https://www.hostinger.com/in/tutorials/how-to-install-ollama) (updated September 2026) describes an **Ubuntu 24.04 with Ollama** template that comes with Ollama, Llama 3 and Open WebUI pre-installed.
For n8n, Hostinger's [self-hosted n8n page](https://www.hostinger.com/self-hosted-n8n) lists three variants of the **Ubuntu 24.04 with n8n** template:
- standard n8n;
- n8n with queue mode, for scaling;
- a version with 100+ pre-made workflows.
## Which plan for which AI job
The key number for local models is RAM. On a CPU-only server, the whole model has to sit in system memory, along with the operating system and any web interface. Ollama's own model library gives the download sizes:
- **Llama 3.1 8B:** [4.9 GB](https://ollama.com/library/llama3.1).
- **Qwen3:** [2.5 GB at 4B, 5.2 GB at 8B, 9.3 GB at 14B and 20 GB at 32B](https://ollama.com/library/qwen3).
| Your project | Suggested plan | Why |
|---|---|---|
| n8n automations calling OpenAI, Claude or Gemini APIs | KVM 1 or KVM 2 | No model runs locally; KVM 2 adds headroom for queue mode and databases |
| Chatbot front end (for example Open WebUI) using an API | KVM 1 or KVM 2 | Light web app, the API does the heavy lifting |
| Small local model, 1B to 4B (for example Qwen3 4B at 2.5 GB) | KVM 2 | Fits in 8 GB with room for the OS |
| 7B or 8B local model (Llama 3.1 8B at 4.9 GB) | **KVM 4** | 8 GB is tight once the OS and web UI load; 16 GB matches Hostinger's own minimum advice |
| 14B local model (Qwen3 14B at 9.3 GB) | KVM 8 | Needs the extra RAM to stay stable |
| 32B model (20 GB) | KVM 8, but slow | Fits in 32 GB, but CPU generation will be very slow |
| 70B model (Llama 3.1 70B at 43 GB) | Not on these plans | Larger than any plan's RAM; use an API or GPU host |
These pairings are our sizing guidance based on the file sizes above, not benchmarks. Hostinger's tutorial recommends **at least 16 GB of RAM and 4 to 8 CPU cores** for an Ollama VPS, which points to KVM 4 or KVM 8. You can [compare Hostinger's KVM 4 and KVM 8 prices for your term here](https://www.hostg.xyz/aff_c?offer_id=815&aff_id=244529).
**On speed:** we have not measured token speeds on these plans, so we will not quote a number. Expect CPU-only generation to be noticeably slower than [ChatGPT](https://chat.openai.com/) or Claude, especially on longer answers. If you want to compare before paying for a server, try a model locally first; our [Ollama vs LM Studio guide](/blog/ollama-vs-lm-studio-local-llms-laptop) shows how.
## Setup: n8n on a Hostinger VPS
1. **Buy a plan.** KVM 2 is the safe choice for most n8n users. [See Hostinger's current KVM 2 price](https://www.hostg.xyz/aff_c?offer_id=815&aff_id=244529).
2. **Choose the server location carefully.** Hostinger's [locations article](https://www.hostinger.com/support/1583267-where-are-your-servers-located) says VPS location is fixed after setup. VPS data centres are in:
- France, Germany, Lithuania and the UK;
- India, Indonesia and Malaysia;
- Phoenix and Boston in the USA;
- Brazil.
3. **Pick the n8n template** (standard, queue mode, or with pre-made workflows) during setup.
4. **Point a domain or subdomain** at the server, so n8n runs over HTTPS on a proper address.
5. **Create the owner account** on first login, with a strong unique password and 2FA.
6. **Store API keys as n8n credentials**, never inside workflow nodes as plain text.
7. **Turn on backups you control.** Weekly VPS backups are included, but also export your workflows regularly.
If you would rather install n8n yourself on a plain Ubuntu server, [n8n's hosting docs](https://docs.n8n.io/hosting/) recommend Docker Compose for production and note that npm installs are deprecated from n8n 3.0.
**Support limits:** Hostinger says its support covers the VPS and keeping n8n running, not troubleshooting your workflows or custom code.
## Setup: Ollama and a private chatbot
1. **Buy KVM 4 or KVM 8** if you want 7B or 8B models.
2. **In hPanel, open your VPS** and choose the **Ubuntu 24.04 with Ollama** operating system template.
3. **Wait for installation**, then connect over SSH to check it is running.
4. **Pull a model that fits your RAM**, for example `ollama pull llama3.1:8b` on KVM 4.
5. **Use Open WebUI** (included in the template) for a ChatGPT-style chat page, and create an admin account straight away.
6. **Keep the Ollama API private.** [Ollama's FAQ](https://docs.ollama.com/faq) says it binds to 127.0.0.1 on port 11434 by default, and that the `OLLAMA_HOST` variable changes this. Do not open port 11434 to the internet. Reach it through Open WebUI's login or an SSH tunnel instead.
7. **Confirm the model is in memory.** The FAQ explains that `ollama ps` shows "100% CPU" when a model is loaded entirely in system memory, which is what you will see on these plans.
On a plain server, the manual route is the install script from [Ollama's Linux docs](https://docs.ollama.com/linux) (`curl -fsSL https://ollama.com/install.sh | sh`), run as a systemd service.
Security matters more on a server you run yourself. Read our pieces on [protecting your AI models and data](/blog/hugging-face-hack-how-to-protect-your-ai-models-and-data) and [securing your AI integrations](/blog/openai-security-breach-lessons-for-protecting-your-ai-integrations) before you connect real customer data.
## When to use an API instead
Self-hosting is not automatically cheaper or better. Use an API (OpenAI, Anthropic, Google and others) when:
- **Quality matters most.** [Frontier models](/blog/ai-safety-talks-between-us-and-china-what-it-means-for-global-tech) are far stronger than any 8B model you can run on a CPU.
- **Speed matters.** Customer-facing chatbots need quick replies; CPU inference may frustrate users.
- **Usage is low or spiky.** Paying per token can cost less than a monthly server you rarely use. Our guide to [cutting your AI API costs](/blog/openai-token-prices-slashed-how-to-cut-your-ai-costs-in-half) shows how to estimate this.
Self-host a model when:
- **Privacy is the priority** and data must stay on your own server;
- **The task is simple** (classification, tagging, short summaries) and a small model is good enough;
- **You want predictable flat costs** for steady background jobs.
**The common middle ground for builders:** run n8n and your chatbot interface on a cheap VPS, and call an API for the model. For more on the trade-off, see our comparison of [open-source and proprietary AI](/blog/open-source-ai-vs-proprietary-which-is-right-for-your-business).
## Downsides to know
- **Renewal prices are higher.** KVM 4 more than doubles from $12.99 to $28.99 a month on renewal.
- **No GPU** on these plans.
- **You are the sysadmin.** Updates, firewalls and backups are your job.
- **Refund limits.** [Hostinger's refund policy](https://www.hostinger.com/legal/refund-policy) allows VPS refunds within 30 days, but only if more than 180 days have passed since your last VPS refund.
## Who it suits, and who should skip it
**Good fit:**
- AI builders running n8n, agents or API-based chatbots who want a fixed monthly cost;
- developers who want a private Ollama sandbox for small models;
- teams in India, Europe or the US who want a nearby server.
**Skip it if:**
- you need GPU speed or models above about 14B;
- you do not want to manage a Linux server;
- you only need a website. Shared hosting is cheaper; see our [Hostinger review](/blog/hostinger-review-2026).
## Verdict
For n8n and API-backed chatbots, a Hostinger KVM 2 is a low-cost, capable base. For local models, budget for KVM 4 or more and accept CPU speeds. Most builders will get the best results from a hybrid: self-host the workflow, rent the intelligence.
**Ready to build?** [Get a Hostinger VPS with the n8n or Ollama template here](https://www.hostg.xyz/aff_c?offer_id=815&aff_id=244529), and use the first 30 days to check that it fits your workload.
Tech-N-AI Talks