# VPS (Virtual Private Server)

> VI: https://5ac.vn/wiki/vps/ · EN: https://5ac.vn/en/wiki/vps/ · lang: en · spec_as_of 2026-10-19 (spec no formal spec) · cluster C17

A virtual machine with its own OS, IP and dedicated resources: where agents, memory, logs and business data live — the sovereignty substrate for SMBs running AI on their own box.

---
## Definition

A VPS (Virtual Private Server) is a slice of a physical machine carved out with its own operating system, its own IP address and a dedicated share of CPU and RAM — it behaves like a dedicated box under your full control, at a fraction of the cost of buying hardware. In 5ac's writing a VPS is not a web-hosting product: it is the self-owned ground an AI system lives on. The sovereignty piece draws the line: when agents, memory, run schedules, access rights and operating logs sit on a VPS or infrastructure the business controls, the model provider becomes just a replaceable component; when all of that lives inside a closed SaaS, the business is renting its operating future.

## Mechanism

The DeepSeek-versus-Workspace-Agents piece arranges the VPS within a three-layer cost path, sequenced by timing rather than by how much hardware you can afford:

| Layer | What it is for | When to move up |
| --- | --- | --- |
| Credits on a closed platform | fast trials of easily reversible use cases | from day one |
| Low-cost APIs | high-volume repetitive work: drafts, classification, summaries | once ROI is proven |
| Private VPS with its own orchestration layer | workflows that touch customer data, quotations, internal SOPs | when data, access control and compliance start to matter |

What resides on the business's VPS: the orchestration layer for the agent team, the knowledge base, recurring schedules, and the operating logs. The precondition is open source — public code means you can run everything on your own server, data never leaves it, and if you one day stop using anyone's service, you keep the whole system. Hosting the models themselves on private GPUs is a separate problem owned by the local LLM deployment entry; this one covers the control layer.

## Application context

For a small business, the VPS is what makes "one person commands a team of agents" operational: a private Ubuntu server holds the entire workflow without depending on any vendor's pricing page or policy change. 5ac's advice follows rhythm, not fashion — do not pay for infrastructure too early, but do not stay trapped behind a single credit-metered gate as volume grows. The trigger to move is a workflow that carries financial consequences.

## Sources fetched

- [AI automation sovereignty for Vietnamese SMBs: use agents without vendor lock-in](/en/blog/ai-automation-sovereignty-vietnam-smb/) — control plane on a VPS: agents, memory, schedules, rights, logs — accessed 2026-10-19
- [Vietnam's Pioneer Bringing the World's #1 Open-Source AI Agent to Enterprise](/en/blog/ai-agent-ma-nguon-mo-viet-nam-tien-phong/) — "run it on your own VPS; data never leaves your server" — accessed 2026-10-19
- [DeepSeek vs Workspace Agents: which agent cost model fits Vietnamese SMBs?](/en/blog/deepseek-vs-workspace-agents-agent-cost-model-smb/) — the three-layer path credits – low-cost API – private VPS, with selection criteria — accessed 2026-10-19

## Related

- [Local LLM Deployment](/en/wiki/local-llm-deployment/)
- [One-Person Company](/en/wiki/one-person-company/)
- [Multi-agent OS](/en/wiki/multi-agent-os/)
- [Vendor lock-in](/en/wiki/vendor-lock-in/)
- [Agent governance](/en/wiki/agent-governance/)