Choose the workload
Decide whether the VPS hosts only your application or also a database, document index, worker queue, or local model.
AI Application Publishing Centre
Publish an app powered by AI.
Your VPS runs the Node.js experience, business logic, and integrations. A connected AI provider runs the model. Configure security, deployment, backups, and monitoring around that clear division.
Before deployment
Build the foundation in this order.
Configure the server
Select Linux and Plesk Web Host Edition, enable Node.js for the domain, then choose a currently supported runtime.
Protect access
Use SSH keys, least-privilege users, environment variables, HTTPS, firewall rules, updates, and rate limits.
Deploy and observe
Set the application root and start command, test a health endpoint, enable restarts, monitor resources, and verify backups.
Choose your setup guide
Different AI ideas need different server plans.
API
2 vCPU / 4 GB starter
AI assistant or content tool
Host the Node.js interface and business logic on your VPS, while a specialized external AI provider performs model inference.
- Store API credentials as protected environment variables.
- Validate requests and limit their size and frequency.
- Stream or queue long responses instead of blocking the web request.
- Keep customer data out of prompts unless the workflow permits it.
RAG
Consider 2 vCPU / 8 GB
Private knowledge search
Retrieve approved documents or database entries before asking the AI service to answer, so responses stay closer to your own material.
- Define which documents are approved for indexing.
- Keep source references with every stored passage.
- Separate public knowledge from account-restricted material.
- Measure memory, index size, and retrieval latency before launch.
JOB
4 GB or more by workload
Automation and background work
Use webhooks and a job queue for classification, summaries, drafting, and other tasks that do not need to finish inside one page request.
- Verify webhook signatures before accepting an event.
- Use a queue with retry limits and failed-job visibility.
- Make repeated events safe to process more than once.
- Record outcomes without logging private prompt content.
LLM
Special sizing required
Locally hosted model
Running model weights on the VPS is not the same as hosting an API-connected app. Useful local models can need much more RAM and often benefit from GPU hardware.
- Choose the model and quantization before choosing hardware.
- Confirm RAM, disk, CPU, and GPU requirements from the model publisher.
- Benchmark tokens per second and concurrent-user behaviour.
- Do not treat the 4 GB starter as a local-model recommendation.
Production checkpoint
Do not launch until these are covered.
- Domain and HTTPS are working
- Secrets are outside source code
- Backups and restore steps are tested
- CPU, RAM, disk, uptime, and logs are monitored
- AI spending and request limits are configured
- Privacy, retention, and human-review rules are documented
Start with enough room to publish properly.
The 2 vCPU / 4 GB VPS is the recommended entry configuration for publishing a small API-powered AI application. Move to 8 GB or higher when the server also carries indexing, databases, multiple applications, or sustained background processing.
Configure the 2 vCPU / 4 GB starter ↗