AI and agents

Ollama: open AI models on your own server

Language models that run on the server's own processor and memory, with a web chat to talk to them, so prompts are not sent to an outside API.

Order a server with Ollama See VPS plans

About Ollama

Ollama is an open-source tool for downloading and running open-weight language models with a single command. Its library includes model families such as Llama, Gemma, Mistral, Qwen and DeepSeek in several sizes, and it exposes a simple HTTP API that other programs can call.

Developers use it to prototype features without per-token costs, and companies use it when text must not leave their own infrastructure, for example internal documents or customer data.

Running it on a server rather than a laptop means the models are available to the whole team and to your applications at any hour. The trade-off is hardware: model size is limited by the server's memory, and generation speed depends on its processor. Models are stored as files on disk, so the disk size also limits how many you can keep.

What people use it for

A chat that never sends data out

Ask questions about sensitive text in the browser chat, knowing the prompt is processed on your server.

A model behind your own app

Call the model from your backend for summaries, classification or extraction, with no API bill per request.

Trying different open models

Pull several models and compare how they handle your real tasks before choosing one.

After you order

Ollama is installed on your server on its first start together with a web chat at http://your-server-address:3000/; port 3000 is opened in the firewall. Sign in with the email and password shown in the Rivervo panel. Models are downloaded on the server: the panel shows the command for that (ollama pull llama3.2 as an example), and the model then appears in the chat.

How it works

Choose Ollama on the server order page and pick a plan with enough memory.
The app is installed on the server on its first start, usually within a few minutes.
Open the server in your panel: the address and the sign-in details are there.

Questions about Ollama

8 GB of memory and 80 GB of disk is the minimum, which comfortably fits small models of a few billion parameters. A model needs roughly as much free memory as its file size, so larger models need a plan with more memory and disk.

Speed depends on the model size and the server's processor cores. Small models answer at a usable pace; big ones are noticeably slower, so start small and move up only if quality requires it.

Yes, it is available on both x86 and ARM plans, which makes ARM a reasonable option for memory-heavy model work.