Ollama on Ubuntu 24.04 LTS — Support & Quick Start

At a glance

Application ports	11434 (HTTP API — no built-in authentication)
Open in browser	http://<VM-IP>:11434/
Sign in as	none — the API has no authentication; restrict port 11434 in the NSG or front it with an authenticating proxy
Service(s)	ollama
Configuration	models in /usr/share/ollama/.ollama/models/; quick-start notes: sudo cat /root/OLLAMA-README.txt
Logs	journalctl -u ollama -f
Version	Ollama 0.31.1 (no models pre-pulled)
Platform	Ubuntu 24.04 LTS

Quick start

1. Deploy from the Azure Marketplace (Get It Now → Create), choosing your SSH key at the Administration step.
2. Allow inbound SSH (22) for yourself plus the application port(s): 11434 (API) — to your clients only; there is no authentication — restrict to your own IP where possible. The in-image firewall already allows them; only the Network Security Group (NSG) keeps them closed.
3. SSH in, `ollama pull llama3.2`, and query the API (see First login below).
4. Point OpenAI-compatible clients at http://<VM-IP>:11434/v1.

First login

1. No credential exists and no model ships (the image is kept small). SSH in and pull one: `ollama pull llama3.2`, then `ollama run llama3.2 "hello"`.
2. API: curl http://localhost:11434/api/generate -d '{"model":"llama3.2","prompt":"Why is the sky blue?"}'; OpenAI-compatible endpoint at http://<VM-IP>:11434/v1.
3. Port 11434 is open in the in-image firewall with NO auth: keep the NSG scoped to your clients or put an authenticating reverse proxy in front. CPU inference works on D-series with small models; resize to an NC/ND GPU size for larger ones.

Still stuck?

Email support@dcassociatesgroup.com (response within 1 business day) or send a message via the contact form. Include the offer name, VM size, region, and any log output — sudo journalctl -u <service> -n 100 usually tells the story.