// Self-hosted AI hosting

AI self-hosted, private and without paying tokens.

Ollama, Open WebUI, Whisper and ComfyUI running on a European VPS. Llama, Qwen, DeepSeek and friends running locally — flat cost, data under GDPR, no corporate filters.

Start with Pulsar Pro See all plans
No per-token billingData in EuropeOpen-weight modelsNo lock-in
// Why self-hosted AI

Flat cost, your data, models without corporate filters.

Self-hosting LLMs changes three things: you stop paying per token, your prompts never leave your disk, and you pick open-weight models with no external policies — including uncensored ones if you need them. PulsarHost gives you the European VPS to do it.

Flat cost vs paying per token

OpenAI and Anthropic bill per input and output token. At 5-10 million tokens/month you easily end up at €100-300/mo. Here you pay for the VPS and run all the machine can handle — unlimited and predictable.

Real privacy, data under GDPR

Your prompts, embeddings and files live only on your NVMe disk. Nothing leaves the VPS — not to OpenAI, not to Anthropic, not to Google. Physical servers in Amsterdam (Tier III), data under European jurisdiction, no Cloud Act.

Open-weight models, the latest

Llama 3.x, Qwen 2.5, Mistral, DeepSeek, Gemma 2, Phi-4. Also uncensored ones like Hermes or Dolphin if you want out of the walled garden of corporate models. Full catalog at ollama.com/library.

// Which plan to pick

Three tiers based on model size.

LLMs on CPU scale with RAM. 8B Q4 fits in 12GB (Pulsar Pro). 70B Q4 fits in 48GB (Pulsar Max ★ — the real sweet spot for serious self-hosted AI). 70B Q5 + ComfyUI batch in 64GB (Pulsar Ultra). Realistic speeds, no marketing fluff.

Pulsar Pro
Llama 8B local · Whisper · embeddings
4 vCPU · 12 GB RAM · 50 GB NVMe
22,99€/mo+ VAT
27,82€ incl. VAT (21 % ES)
  • Llama 3 8B Q4 at 8-15 tokens/s
  • Open WebUI ChatGPT-style included
  • Whisper transcription + light RAG
Get started
★ Recommended
Pulsar Max
Llama 70B Q4 · Open WebUI · serious RAG
12 vCPU · 48 GB RAM · 125 GB NVMe
67,99€/mo+ VAT
82,27€ incl. VAT (21 % ES)
  • Llama 3.1 70B Q4 fits comfortably (48GB)
  • Quality close to GPT-4 without token cost
  • Async/batch at 1-3 t/s · ideal RAG and summaries
Get started
Pulsar Ultra
70B Q5 + ComfyUI batch · multi-model
16 vCPU · 64 GB RAM · 200 GB NVMe
97,99€/mo+ VAT
118,57€ incl. VAT (21 % ES)
  • 70B Q5 with room for long context
  • ComfyUI / SD batch (not interactive)
  • Multi-model orchestration + workers
Get started
// AI Self-Hosted FAQ

What's realistic, no overpromising.

Self-hosting AI is great — but it's worth knowing what works well on CPU and what needs GPU. Answers here without smoke. Don't see yours? Email info@pulsarhost.es.

Is it realistic to run LLMs on CPU without GPU?
Depends on the model and use. Llama 3 8B Q4 runs at 8-15 tokens/s on Pulsar Pro/Max — viable for personal chat and batch. Llama 70B Q4 at 1-3 tokens/s on Pulsar Max/Ultra — viable for async (summaries, RAG, batch generation). For fluid interactive chat at GPT-4 speed you need GPU; PulsarHost is CPU-only today. If your case needs GPU, email us and we'll guide.
What privacy do I have vs ChatGPT or OpenAI?
Your prompts, embeddings and files live only on your NVMe disk. Nothing leaves the VPS — not to OpenAI, not to Anthropic, not to Google. Physical servers in Amsterdam (Tier III), data under European GDPR, no Cloud Act. You can run uncensored models (Hermes, Dolphin) without external policies. You decide what comes in and what goes out.
What models does Ollama support?
Any GGUF — Llama 3.x (8B / 70B / 405B), Qwen 2.5 (7B/14B/32B/72B), Mistral, DeepSeek (V2/V3, Coder), Gemma 2, Phi-4, uncensored Hermes/Dolphin. Switch model with `ollama pull <name>`. Full catalog at ollama.com/library. Open WebUI gives you a ChatGPT-style UI for everything.
Does ComfyUI or Stable Diffusion work?
Yes, but on CPU it's slow — seconds to minutes per image depending on resolution and model. Useful for batch jobs or light LoRA training; not recommended for interactive use. If you're coming to generate real-time images Midjourney/DALL-E style, you need GPU (we don't offer dedicated GPU yet).

Your AI, on your VPS, no token counters.

Pulsar Pro from €22.99/mo. Local Llama 8B · Open WebUI · Whisper. No lock-in, data in Europe.

Start now