Run your own AI stack

Your own AI.
No GPU required.

Most self-hosted AI is text and automation, and it runs fine on a normal CPU VPS. Here is exactly what you can run on ours, what runs slower, and what needs a GPU we do not pretend to sell.

What runs on each plan Browse the tools
Straight answers first

What a CPU VPS is great at, and what it is not.

We would rather tell you the truth than sell you the wrong plan.

Runs great on our VPS

Self-hosted chat with open models up to about 8B (Ollama plus Open WebUI), AI automation with n8n, document search and RAG with Qdrant, and batch speech-to-text with Whisper. This is the majority of what people self-host.

Needs a GPU, which we do not offer yet

Image generation (Stable Diffusion, ComfyUI, Fooocus), high-throughput production serving with vLLM, real-time voice, and very large models above roughly 14B at usable speed. If that is your workload, a CPU VPS is the wrong tool and we will say so.

The open-source AI catalog

Every tool you might run, labelled honestly.

Open-source, self-hosted, and yours. Each card tells you how it behaves on a CPU VPS.

Runs great Runs, slower Needs a GPU

Run the models

Serve open LLMs locally.

Ollama

Runs great

One command to download and serve open models over an OpenAI-compatible API. The easiest place to start.

llama.cpp

Runs great

The lean inference engine under Ollama. The most memory-efficient way to run GGUF models.

LocalAI

Runs great

A drop-in OpenAI-compatible API for local text models, so your existing code just works.

GPT4All

Runs great

Simple local model chat that was built to run comfortably on CPU.

vLLM

Needs a GPU

High-throughput production serving. Requires an NVIDIA GPU, so not on our CPU plans.

text-generation-webui

Needs a GPU

A power-user model UI. Useful model sizes really want a GPU behind them.

Your private ChatGPT

Chat front ends for your models.

Open WebUI

Runs great

The popular self-hosted ChatGPT-style interface for Ollama. The UI is light. The model does the work.

LibreChat

Runs great

Multi-provider chat that can route to your local models and hosted APIs side by side.

AnythingLLM

Runs great

Document question-answering and RAG in one clean workspace.

Lobe Chat

Runs great

A polished, mobile-installable chat UI with a plugin marketplace.

Automate with AI

Workflows and agents.

n8n

Runs great

Self-hosted workflow automation with native AI nodes. One of the best fits for a small VPS.

Flowise

Runs great

Drag-and-drop builder for LLM apps and agents.

Dify

Runs great

Build and run LLM apps with RAG and observability built in.

Langflow

Runs great

Visual builder for LangChain and agent flows.

Activepieces

Runs great

Open-source automation with an AI copilot for building flows.

Memory and search

RAG and vector databases.

Qdrant

Runs great

Fast Rust vector database. The easiest self-hosted starting point for RAG.

Chroma

Runs great

Lightweight embeddings store, ideal for prototypes and smaller datasets.

pgvector

Runs great

Vector search inside a PostgreSQL database you may already be running.

Weaviate

Runs great

Vector database with strong hybrid keyword-and-vector search.

Milvus

Runs, slower

Built for very large scale. Heavier than a single VPS needs for most projects.

Voice

Speech to text and text to speech.

faster-whisper

Runs great

Fast, low-memory speech-to-text for batch transcription jobs.

Piper

Runs great

Lightweight text-to-speech built for low-resource machines.

Coqui XTTS

Runs, slower

Voice cloning that works on CPU if you are patient. Faster with a GPU.

Images

These need a GPU.

Stable Diffusion

Needs a GPU

Image generation with the Automatic1111 web UI. Not practical on CPU.

ComfyUI

Needs a GPU

Node-based image workflows. GPU required for usable speeds.

Fooocus

Needs a GPU

Simplified image generation. GPU required.

Deploy and store

Tools people run alongside AI.

Coolify

Runs great

Self-hosted deploy platform with one-click templates for Ollama, n8n and more.

Supabase

Runs great

Open-source Postgres backend that ships with pgvector for RAG.

Nextcloud

Runs great

Self-hosted files and productivity that can call your local AI.

Match the model to the memory

What runs on each VPS size.

RAM is what matters most for AI workloads. Pick a plan with enough headroom for what you want to run. For anything Docker-based, choose a KVM plan.

RAMNameMarshal planWhat runs comfortably
4 GB OVZ 4 GB $18.99/mo
KVM 4 GB $24.99/mo		 n8n, Open WebUI, Coolify basics, self-hosted Supabase, and small models like Llama 3.2 1B and 3B, Phi-3 Mini, plus Piper text-to-speech and batch Whisper (base or small).
8 GB OVZ 8 GB $32.99/mo
KVM 8 GB $41.99/mo		 The sweet spot. Mistral 7B, Llama 3.1 8B or Qwen 2.5 7B through Ollama at roughly 4 to 15 tokens per second, production n8n, Qdrant at around one million vectors, AnythingLLM and LibreChat.
16 GB OVZ 16 GB $54.99/mo
KVM 16 GB $69.99/mo		 Qwen 2.5 14B and Gemma 2 9B, two models loaded at once, Weaviate, larger n8n plus Postgres stacks, and faster-whisper large-v3 for batch transcription.
32 GB KVM 32 GB $119.99/mo Mixtral 8x7B, higher-quality quantization of 7B to 13B models, and multi-service Docker Compose stacks running several tools together.
Start on KVM 8 GB See all VPS plans

Token-per-second figures are typical CPU ranges for guidance, not a guarantee. Actual speed depends on the model, quantization and the specific workload.

Pick the right VPS type

OpenVZ or KVM for AI?

Short version: if you are installing tools with Docker, and most of these use Docker, choose KVM.

Recommended for AI

KVM VPS

Full hardware-level virtualization. Runs Docker, custom kernels and container stacks without fuss, which is what almost every AI tool here expects.

  • Best for Ollama, Open WebUI, n8n and Docker Compose stacks
  • Full control over the operating system
  • From $24.99/mo at 4 GB, up to 32 GB
Choose a KVM plan
Lighter workloads

OpenVZ VPS

A lightweight container VPS that is cheaper and quick to spin up. Great for simpler setups, but it is not the right home for a Docker-based AI stack.

  • Good for lightweight scripts and simple services
  • Lower cost per gigabyte of RAM
  • From $18.99/mo at 4 GB, up to 16 GB
Choose an OpenVZ plan
Common questions

Self-hosted AI, answered plainly.

Can I run a ChatGPT-style AI without a GPU?
Yes. Text models in the 1B to 8B range run on a normal CPU VPS. Pair Ollama with Open WebUI and you have your own private chat interface. You do not need a GPU for this. A GPU only matters for large models at high speed or for image generation.
How much RAM do I need to run Ollama?
Plan for about 8 GB of RAM to run a 7B or 8B model like Mistral 7B or Llama 3.1 8B comfortably. Smaller 1B to 3B models run in 2 to 4 GB. More RAM lets you load bigger models or keep several loaded at once.
What is the best open-source LLM for a small VPS?
For an 8 GB VPS, Mistral 7B and Llama 3.1 8B are the mainstream picks. Phi-3 Mini and Llama 3.2 3B are faster and fit smaller plans. Expect roughly 4 to 15 tokens per second on CPU, which is fine for personal use, internal tools and low-traffic APIs.
Can I run Stable Diffusion on a VPS without a GPU?
Not practically. Image generation with Stable Diffusion, ComfyUI or Fooocus needs a GPU to be usable. On a CPU it can take minutes per image, so we do not recommend it on our CPU plans and we will not pretend otherwise.
Is CPU inference fast enough for real use?
For 7B to 8B models, expect about 4 to 15 tokens per second on a CPU VPS. That is well suited to personal assistants, background jobs, automation and low-concurrency APIs. It is not built for high-traffic real-time chat serving many users at once.
Should I choose OpenVZ or KVM for AI tools?
Choose KVM. Most of these tools install with Docker, and Docker needs the full virtualization that KVM provides. OpenVZ is a lighter container VPS that is great for simple workloads but is not the right fit for a Docker-based AI stack.

New guide: install Ollama on a VPS

Step by step, from a fresh VPS to your own LLM API. More on the blog.

Read the guide

Ready to host your own AI?

Start on a KVM VPS with the RAM you need. Free SSL, daily backups and real support are included, same as every NameMarshal plan.

Get started