Ollama on AWS gives you a private ChatGPT-style setup: Ollama runs open models on the server, and OpenWebUI gives your team a browser interface for chatting with them, uploading documents and sharing prompts. Meetrix's AMI installs both, Open WebUI 0.9.5 and Ollama 0.23.3 on Ubuntu 24.04, behind HTTPS, in your own AWS account.
One thing to settle before you launch: the instance type. It decides which models are usable, and that matters more than any setting in the UI.
Quick Video Guide
OpenWebUI AWS setup: what the AMI includes
- OpenWebUI and Ollama installed and connected on one EC2 instance (Ollama listens on its default port 11434, inside the instance)
- HTTPS with a Let's Encrypt certificate, automatic when your domain is on Route 53
- No bundled model; you pull the ones you want after launch
- A 5-day free trial, then an hourly software fee on your AWS bill alongside EC2
Deploy Ollama on AWS in six steps
- Open the OpenWebUI listing on AWS Marketplace, subscribe and accept the terms.
- Choose "CloudFormation Template" and your region, then launch the stack.
- Fill in
Admin Email, your domain andInstanceType(t3.medium is the minimum; see below). A vCPU error means a vCPU quota increase first. - Wait 5-10 minutes, copy
PublicIpfrom the Outputs tab and point your domain's A record at it. - Open
DashboardUrland create the admin account straight away. The first account to sign up becomes the admin. A 502 in the first few minutes just means it's still starting. - In the admin settings, pull a model, then start chatting.
Hosting Ollama on AWS: CPU or GPU?
The Meetrix listing offers CPU instances, general purpose, compute optimized and memory optimized, from medium to xlarge. That's a real constraint, and it's better to know it up front than after a slow first answer.
- On CPU, think small and quantized. Models of roughly 1B to 8B parameters in 4-bit quantization run at a usable pace for a handful of people. Memory is the hard limit: the model has to fit in RAM with room to spare, which is why a memory-optimized r6a.large often beats a compute-optimized instance of the same price.
- Bigger models want a GPU. A 70B model on CPU technically runs and practically doesn't help anyone. For that, or for dozens of concurrent users, look at a GPU instance and a serving engine built for throughput. Our Ollama vs vLLM comparison explains where each one fits.
- Or skip local models for some tasks. OpenWebUI also talks to OpenAI-compatible APIs, so you can keep sensitive chats on a local model and send the rest to a hosted one.
Not sure which model to start with? Our guide to the best self-hosted LLM families covers the trade-offs.
OpenWebUI S3 storage
By default, files people upload for RAG live on the instance's disk. OpenWebUI can store them in S3 instead: set STORAGE_PROVIDER to s3 and add the bucket name, region and credentials, as described in OpenWebUI's S3 storage guide. That keeps uploads safe if you replace the instance, and on EC2 it's tidier to grant access through the instance's IAM role than to paste access keys.
Who it's for
Teams that want a ChatGPT-style assistant without sending prompts and documents to a third party: internal knowledge-base search over your own files, drafting and summarizing for staff who handle client data, or a sandbox for developers to try open models before committing to one. OpenWebUI's license changed in 2025 to add a branding clause; our OpenWebUI vs LibreChat comparison covers what that means if you plan to rebrand it.
Frequently Asked Questions
How do I deploy Ollama on AWS?
The quickest way is Meetrix's OpenWebUI and Ollama AMI: subscribe on AWS Marketplace, launch the CloudFormation stack with your domain and instance type, then open the DashboardUrl and create the admin account. Then pull a model from the OpenWebUI admin settings.
Does Ollama on AWS need a GPU?
No, but it decides what's practical. The Meetrix listing offers CPU instances, where small quantized models (roughly 1B to 8B parameters) answer at a usable pace and larger ones get slow. For bigger models or many users, a GPU instance with vLLM is the better fit.
Which models can I run with Ollama?
Any model in the Ollama library, such as Llama, Mistral, Gemma, Qwen and DeepSeek variants, plus your own GGUF files. On a CPU instance, pick small, quantized versions and check that the model fits in RAM with room to spare.
Can OpenWebUI store files in S3?
Yes. Set STORAGE_PROVIDER to s3 and add the bucket name, region and credentials or endpoint in OpenWebUI's environment. Uploaded documents then go to S3 instead of the instance's disk.
Who creates the OpenWebUI admin account?
The first person to sign up after launch becomes the admin, so open the DashboardUrl and register right away. Further users are created or approved from the admin panel.
Can I use this in a private VPC?
Yes. The stack runs in your own AWS account, and nothing in a chat leaves the instance unless you connect an external API provider. Restrict the security group to your office or VPN range if only staff should reach it.
Do I need technical experience to deploy this?
Not much. Follow the step-by-step Developer Guide to launch and run it. You'll need an AWS account, a domain and basic comfort with the AWS console.
Ready to take control of your AI infrastructure?
Deploy OpenWebUI and Ollama on AWS with Meetrix for a secure, scalable, and fully-supported solution. Get started in minutes and unlock the full potential of self-hosted large language models.
Launch your private LLM interface today
