Agents and crews
LangChain, CrewAI, AutoGen, Letta and Claude Code-style agents, running as long-lived processes that call a hosted model.
Use case · AI agents
AI agent hosting: run AI agents on a VPS, no GPU
Yes — you can run an AI agent on a VPS with no GPU, because the model runs through your own OpenAI or Anthropic API key while the server runs the agent. An AI agent VPS is a Linux or Windows server where your agents, workflows and tools stay online 24/7 as systemd services or Docker containers, with full root or Administrator access to install whatever your stack needs.
- Model runs via your own API key
- No GPU needed to run agents
- Linux or Windows, full control
What you can run on an AI agent VPS
Workflows and MCP servers
n8n automations, MCP servers and schedulers that call LLM APIs and move data between your own apps.
RAG and vector search
Qdrant, Chroma or pgvector holding your embeddings on disk, queried by the agent on the same server.
Do you need a GPU to run an AI agent?
No. An AI agent needs a GPU only when the large language model runs on the same machine, and most agents do not run the model themselves. Frameworks like LangChain, CrewAI and n8n send each prompt to a hosted model over an API — OpenAI, Anthropic, Google or any OpenAI-compatible endpoint — and spend their own time on orchestration: routing steps, calling tools, parsing results and waiting on the network. That work is light on CPU and memory, so a CPU VPS — part of our AI VPS range — runs it well around the clock. The GPU, and the cost of running the model, stay on the API provider's side: you pay us for the always-on server and the provider for the tokens you use.
A small quantized model can run locally on the CPU with a tool like Ollama or llama.cpp, but it is slow and limited to small models — fine for a test or a lightweight classifier, not for fast or large-model inference. For production answers, keep the model on a hosted API and let the VPS run the agent around it.
What an AI agent VPS is good for, and what it is not
| Task | Good fit? | Why |
|---|---|---|
| Running an AI agent or crew 24/7 | Yes | Orchestration is light CPU work; the model runs on an API |
| n8n and MCP servers that call LLMs | Yes | Node and Docker services that stay online, with full root to install them |
| A vector database for RAG | Yes | Similarity search runs on the CPU and in RAM, with no GPU |
| Training or fine-tuning a model | No | Needs a GPU server, which we do not sell |
| Image or video generation locally | No | GPU-only work; call a hosted API for it instead |
| Fast local inference of a large model | No | There is no GPU, and a small model on the CPU is slow |
Which plan for your AI agent
Shown in Dallas. Because the model runs on an API, size the plan by what your agent and its tools keep in memory, not by GPU. T1 runs one lightweight agent; T2 runs an agent with Docker and a small database; T3 runs several agents at once, or one agent beside a vector database.
T1 · Dallas
AMD Ryzen
$14.99/mo
USD per month · billed for 1, 3, 6 or 12 months
- 2 shared vCPU
- 2 GB DDR5 RAM
- 40 GB Gen4 NVMe SSD
- 1 Gbps port · 1 IPv4
An agent with Docker and a database
T2 · Dallas
AMD Ryzen
$24.99/mo
USD per month · billed for 1, 3, 6 or 12 months
- 4 shared vCPU
- 4 GB DDR5 RAM
- 80 GB Gen4 NVMe SSD
- 1 Gbps port · 1 IPv4
T3 · Dallas
AMD Ryzen
$39.99/mo
USD per month · billed for 1, 3, 6 or 12 months
- 6 shared vCPU
- 8 GB DDR5 RAM
- 160 GB Gen4 NVMe SSD
- 1 Gbps port · 1 IPv4
Plan prices do not go up at renewal, and there is no setup fee. Prices and specs from our live catalog, checked .
Run an AI agent 24/7 on Linux (systemd or Docker)
Our Linux VPS is live — Ubuntu, Debian or AlmaLinux, reached as root over SSH — and it is the simplest home for an always-on agent (Windows Server is available too if your tooling needs it). An agent you start in an SSH session stops when you log out, so run it as a service that restarts on its own:
- systemd runs your agent as a managed service: write a unit that runs
python -m your_agentornode agent.js, setRestart=always, thensystemctl enable --now your-agent. It starts on boot and comes back if the agent crashes. - Docker packages the agent and its dependencies;
docker run -d --restart=unless-stoppedkeeps the container alive across reboots. Install Docker from your distribution's packages. - n8n installs with npm or runs as a Docker container — our n8n hosting walks through the same setup — and you keep its editor behind a firewall rule or a reverse proxy rather than open to the internet.
You install these yourself with apt, npm, pip or Docker — you have full root, so any runtime or tool that runs on Linux works.
Keep your model API keys safe on the server
An agent needs its model API key and any tool credentials at runtime, and those should never sit in the code you commit or share. On the server:
- Put keys in environment variables or a
.envfile that only the service user can read (chmod 600), and load them at start — never hard-code them in the script. - For systemd, use
EnvironmentFile=pointing at a root-owned file, or theLoadCredential=mechanism, so keys stay out of the unit file and the process list. - Close every port you do not need. Reach the server over SSH, keep agent dashboards such as n8n behind the firewall or a VPN, and expose only what the public actually needs.
- A hosted model key bills by usage, so treat a leaked key as money: rotate it at the provider straight away if a server is ever compromised.
Vector database and RAG on a CPU VPS
Retrieval-augmented generation (RAG) stores your documents as embeddings in a vector database and pulls the closest matches to feed the model as context. Both the database and the search run on the CPU: Qdrant, Chroma and pgvector (the PostgreSQL extension) all run well without a GPU, because similarity search over stored vectors is CPU and memory work, not model inference. Create the embeddings through the same hosted API your agent uses, or with a small local embedding model when latency does not matter.
Memory is what grows with the data. Each vector is a few kilobytes, so a few thousand documents sit comfortably on T2, while hundreds of thousands of chunks want T3 or higher to keep the index in RAM. Gen4 NVMe SSD storage keeps loading and persistence fast. Install the database yourself with Docker or your distribution's packages.
AI agent hosting guides
Workloads
How to make a Minecraft server on a Linux VPS, step by step
Make a Minecraft server on a Linux VPS: install Java and Paper, accept the EULA, edit server.properties, open port 25565, run it with systemd, and connect.
RDP how-to
MSTSC command guide: every switch with examples
A complete guide to the mstsc command: /v, /admin, /f, /w and /h, /multimon, /span, /public and /shadow, with copy-ready examples for Windows Server RDP.
RDP how-to
RDP port 3389 explained: TCP vs UDP and when to change it
Remote Desktop uses TCP and UDP port 3389 by default. See how RDP uses each, how to check the port, and how to change it safely on Windows Server.
AI agent hosting FAQ
Do I need a GPU to run an AI agent?
No. The agent sends each prompt to a hosted model over an API, so the GPU work happens at the provider. The VPS runs the orchestration, tools and schedules, which are light CPU and memory work. You only need a GPU if you insist on running a large model on the server itself.
Can I run an AI agent on a VPS?
Yes. Run LangChain, CrewAI, AutoGen, Letta or your own agent as a systemd service or a Docker container so it stays online 24/7 and restarts after a crash or reboot. You have full root, so you install the runtimes and packages it needs.
How much RAM does an AI agent need?
Usually tens to a few hundred megabytes per agent plus its Python or Node runtime, so T1 (2 vCPU, 2 GB RAM and 40 GB Gen4 NVMe SSD) runs one lightweight agent. T2 (4 vCPU, 4 GB RAM and 80 GB Gen4 NVMe SSD) gives room for an agent with Docker and a small database; T3 (6 vCPU, 8 GB RAM and 160 GB Gen4 NVMe SSD) runs several agents or a vector database.
How much RAM does n8n need?
n8n itself runs in a few hundred megabytes. With its Docker image and a PostgreSQL database alongside it, give it T2 (4 vCPU, 4 GB RAM and 80 GB Gen4 NVMe SSD) or more; many parallel workflows and large payloads want T3.
Can I run a local LLM on this VPS?
Only a small quantized model, through a tool like Ollama or llama.cpp, and it runs slowly on the CPU because there is no GPU. It suits a test or a lightweight classifier; for fast or large-model answers, call a hosted API from the agent instead.
Can I run a vector database like Qdrant, Chroma or pgvector for RAG?
Yes. Vector search runs on the CPU and in RAM without a GPU, so Qdrant, Chroma and pgvector all work. Size the tier by how many vectors you hold: a few thousand documents fit T2, hundreds of thousands want T3 or higher to keep the index in memory.
How do I keep an agent running after I close my SSH session?
Run it as a service. A systemd unit with Restart=always, or a Docker container with --restart=unless-stopped, keeps the agent alive through disconnects, crashes and reboots.
Where do I store my API keys?
In environment variables or a file only the service user can read, loaded at start — never hard-coded in the script. For systemd, use EnvironmentFile= with a root-owned file so the key stays out of the unit and the process list, and rotate any key that is exposed.
Deploy your AI agent server
A CPU VPS that runs your agents 24/7 — no GPU, your own API key. Plans from $14.99/mo.