Hostinger Server Setup
// Create VPS · Install Docker · Configure network & firewall
| Resource | Hostinger KVM 8 | Why |
|---|---|---|
| vCPU | 8 cores AMD EPYC | Parallel container workloads |
| RAM | 32 GB | Sim.ai alone needs 12 GB |
| Storage | 400 GB NVMe | LLM models + DB growth |
| OS | Ubuntu 24.04 LTS (Noble Numbat) | cgroup v2 · best Docker support |
1
Create Your Hostinger VPS
▲
Go to: hostinger.com/vps-hosting → Choose Plan → KVM 8
During setup select Ubuntu 24.04 LTS as the OS. After provisioning, note your server's public IP from the hPanel dashboard.
Hostinger provisions in under 60 seconds. You'll have hPanel access immediately with a built-in Docker Manager GUI included for free.
2
SSH Into Your Server from Mac
▼
bash · your Mac terminal
ssh root@YOUR_SERVER_IP # Verify Ubuntu version lsb_release -a # Expected: Ubuntu 24.04.x LTS # Confirm cgroup v2 (Ubuntu 24.04 default) stat -fc %T /sys/fs/cgroup # Expected output: cgroup2fs ✓
Replace before running:
| Placeholder | What to put | Where to find it |
|---|---|---|
| YOUR_SERVER_IP | Your Hostinger server's public IP address | hPanel → VPS section → click your server → IP shown at top of dashboard (e.g. 82.115.x.x) |
3
Update System & Install Essentials
▼
bash
apt update && apt upgrade -y apt install -y curl wget git nano ufw fail2ban htop
4
Install Docker & Docker Compose
▼
bash
# Official Docker install — works perfectly on Ubuntu 24.04 curl -fsSL https://get.docker.com -o get-docker.sh sh get-docker.sh apt install -y docker-compose-plugin # Verify docker --version docker compose version
Ubuntu 24.04 ships cgroup v2 by default — Docker detects it automatically. No manual configuration needed.
5
Create Docker Network & Folder Structure
▼
One shared Docker bridge network lets all containers talk to each other by name — completely internal to the server, no extra Hostinger networking needed. Run each command separately:
bash · command 1 of 3
docker network create ai-stack
bash · command 2 of 3
mkdir -p ~/ai-stack/{ollama,claude-code,openclaw,sim}
bash · command 3 of 3
echo "✅ Network and folders ready"
6
Configure Firewall (UFW)
▼
Run each command separately:
bash · command 1 of 5
ufw allow OpenSSH
bash · command 2 of 5
ufw allow 8080
bash · command 3 of 5
ufw allow 3000
bash · command 4 of 5
ufw allow 11434
bash · command 5 of 5 — enable & verify
ufw --force enable && ufw status
Restrict Ollama to your IP only in production:
ufw allow from YOUR_IP to any port 11434Claude CLI (host install)
// Anthropic's official CLI on the host · used directly via SSH / web terminal
Two ways to use Claude Code on this stack. This tab covers installing the official
claude CLI directly on the host (Ubuntu), so you can drop into claude from any SSH session or the web terminal. Tab 4 (Claude Code Container) covers a containerized variant with isolated env vars per project. Most people install both: the host CLI for ad-hoc work, the container for repeatable per-project runs.Requires an Anthropic account (sign-in flow) or an API key from console.anthropic.com. The CLI prompts you on first run.
1
Install the CLI
▲
Anthropic ships an installer that drops a single binary under ~/.claude/bin/ and adds it to your PATH on next shell. No package manager pollution.
bash · one-line install
curl -fsSL https://claude.ai/install.sh | bash
Re-source ~/.bashrc (or open a new shell) so the new PATH takes effect:
bash · reload shell
source ~/.bashrc
Alternative: npm. If you already have Node 20+ on the host,
npm install -g @anthropic-ai/claude-code also works. The official installer is recommended because it self-updates and stays isolated under ~/.claude/.2
Verify the install
▼
bash · expect a version like 1.x.y or claude-code 0.x
claude --version
bash · confirm the binary location
which claude
3
Sign in (one-time)
▼
First invocation opens a browser sign-in flow (paste the printed URL into your local browser since the VPS has no GUI). Subsequent runs reuse the saved credentials at ~/.claude/.
bash · interactive sign-in on first run
claude
API-key-only mode (no browser, useful for CI / automation): export
ANTHROPIC_API_KEY before invoking the CLI. secrets/secrets.env has a placeholder; populate it, source it, then run claude.4
Smoke test
▼
Drop into a project directory and ask Claude to summarise it:
bash · ad-hoc question from inside the project tree
cd ~/pocketcode-project claude
At the prompt: "summarise this repo in five bullets". If you get a sensible answer, you're done — close with Ctrl+D or /exit.
Official docs: docs.claude.com/en/docs/claude-code · Updates:
claude update · Uninstall: delete ~/.claude/Ollama
// Local LLM runner · CPU mode · Port 11434 · ollama/ollama
Docs: docs.ollama.com/docker · Image: hub.docker.com/r/ollama/ollama
Running in CPU mode on your Hostinger KVM 8. Inference is slower than GPU but fully functional — small models (3B–8B) respond in 5–20 seconds which is fine for personal use and curriculum development.
1
Run Ollama Container (CPU mode)
▲
This is a multi-line command. Create a script file, paste the content, make it executable, then run it:
bash · step 1 — create the script file
touch ~/ai-stack/ollama/run-ollama.sh
bash · step 2 — open in nano editor
nano ~/ai-stack/ollama/run-ollama.sh
bash · step 3 — paste this into nano, then press Ctrl+O to save, Ctrl+X to exit
#!/bin/bash docker run -d \ --name ollama \ --network ai-stack \ --restart unless-stopped \ -v ollama-data:/root/.ollama \ -p 11434:11434 \ ollama/ollama
bash · step 4 — make executable and run
chmod +x ~/ai-stack/ollama/run-ollama.sh
bash · step 5 — execute the script
bash ~/ai-stack/ollama/run-ollama.sh
2
Pull Models & Test
▼
bash · command 1 — pull a model
docker exec -it ollama ollama pull llama3.2
bash · command 2 — test the API
curl http://localhost:11434/api/generate -d '{
"model": "llama3.2",
"prompt": "Hello World!"
}'
bash · command 3 — list downloaded models
docker exec -it ollama ollama list
Other containers reach Ollama at
http://ollama:11434 via the ai-stack network.3
Recommended Models by VRAM
▼
| Model | VRAM | Best for |
|---|---|---|
| llama3.2:3b | ~2 GB | Fast responses, low resource |
| llama3.2:8b | ~5 GB | Good quality general use |
| mistral:7b | ~5 GB | Coding + reasoning |
| deepseek-r1:8b | ~6 GB | Strong reasoning tasks |
| gemma2:9b | ~7 GB | Curriculum & education content |
