PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteYou can run Microsoft’s Phi-3 Mini privately on a Linux VPS by pairing Ollama (the model runtime and API) with Open WebUI (the browser interface). Install the phi3 model in Ollama, put both services on a private Docker network, persist their data in named volumes, and expose only Open WebUI through an HTTPS reverse proxy. Phi-3 Mini is small enough for CPU inference, while a GPU can improve latency for concurrent users.
Antihost offers Linux VPS plans with full root access, unlimited traffic on a 1 Gbps port and a 14-day free trial.
This guide builds that layout, explains when to choose CPU or GPU, shows the exact connection settings, and covers the failures that commonly produce an empty model selector or a stalled chat.
As an Amazon Associate I earn from qualifying purchases.
Free tools Windows power users keep installed
One-click scans. No signup required.
What you are building
The finished deployment has four parts:
- Ollama: downloads, loads and serves Phi-3 through an HTTP API on port 11434.
- Phi-3 Mini: the
phi3model installed in Ollama. Ollama’s current quick-start lists 3.8 billion parameters and a 2.3 GB download. - Open WebUI: the web application where users select models and hold chats. Its official quick-start states, “Open WebUI has no models of its own.”
- Reverse proxy and firewall: terminate HTTPS, authenticate users and keep Ollama’s API off the public Internet.
Open WebUI does not download a model for you. The model must exist in Ollama and Open WebUI must be able to reach Ollama’s API. A browser connects to Open WebUI; it should not connect directly to port 11434.
Phi-3 Mini capabilities and realistic expectations
The Phi-3 technical report describes Phi-3 Mini as a 3.8-billion-parameter model trained on 3.3 trillion tokens. The report records 69% on MMLU and 8.38 on MT-Bench. Those are published evaluation results, not a speed guarantee for your VPS: actual throughput depends on CPU or GPU, context length, quantization, storage and the number of simultaneous generations.
#1 Best Overall
- BLUE LIGHT BLOCKING GLASSES: Our blue light glasses for women boast UV400 protection, blocking harmful blue rays from computer screens. With these bluelight glasses women, you can enjoy gaming, reading, or watching TV with no worries of eye strain.
- EYE RELIEF: Our blue light glasses are engineered to reduce blurry vision or strain caused by prolonged screen time. These womens blue light glasses aid in better sleep by preventing blue light from disrupting your natural sleep cycle, helping you fall asleep faster and enjoy a more restful night's sleep, making them the perfect christmas present.
- COMFORTABLE: Crafted from durable material, our trendy blue light glasses for woman or blue light glasses men ensure long-lasting comfort. The high-transmittance lenses accessories, gamer gifts for men, help keep your eyes relaxed. Perfect stocking stuffers, gifts for gamers men, christmas gifts for men, halloween costumes.
- LIGHTWEIGHT: These clear blue light glasses feature a lightweight frame, meaning they’ll stay comfy for long-term wear. Perfect as blue blockers glasses for women or mens blue light glasses, these non-prescription fake glasses for men can be worn throughout the day. The blue light blocking glasses women feature polycarbonate HD lenses, which are anti-reflective and restore true color.
- WORRY-FREE PURCHASE: If you experience any issues, our support team is ready to assist you quickly. These are the perfect costume glasses for cosplay dressing up.
The 2.3 GB figure is the Ollama download size listed for phi3. Your deployment also needs operating-system space, Docker layers, logs, temporary files and room for future model versions. Leave free disk capacity rather than provisioning a volume that is only slightly larger than the download.
Choose a VPS layout before installing
Separate Ollama and Open WebUI containers
Two services are the most flexible arrangement. Each container can be upgraded independently, Ollama’s model volume can be backed up separately, and the Docker network gives Open WebUI a stable service name. This is the layout used in the compose example below.
All-in-one Open WebUI image
Open WebUI also publishes an :ollama image that bundles an Ollama process with the interface. It is convenient for a single-user trial, but a combined container couples upgrades and makes it harder to isolate the model API. Use it when simplicity matters more than independent lifecycle control.
Ollama on the host, WebUI in Docker
This can work when Ollama is already managed by the VPS operating system. The container must resolve the host through Docker’s gateway mapping and Ollama must listen on an interface visible to that mapping. Do not bind port 11434 to every public interface merely to make this work; use a private interface or firewall rule.
| Decision | Recommended default | When another choice fits |
|---|---|---|
| Inference hardware | CPU VPS for one or a few users | GPU VPS when concurrent requests or response latency justify the added cost |
| Container layout | Separate Ollama and Open WebUI services | :ollama image for a disposable, simple installation |
| Network exposure | Publish WebUI only; keep 11434 private | Expose Ollama only for a deliberately authenticated remote-client design |
| Data storage | Named volumes with scheduled backups | Bind mounts when your backup and permission policy requires host paths |
| Public access | Reverse proxy with HTTPS and authentication | Private VPN access for a small trusted group |
VPS sizing without a misleading minimum
The available documentation does not establish a universal minimum RAM or vCPU count for Phi-3 Mini. Parameter count alone cannot predict whether a deployment will be responsive or run out of memory. Size the VPS from the workload you actually expect:
- Context length: longer prompts and conversation histories increase memory use and generation time.
- Concurrency: each simultaneous generation competes for CPU, memory and, on a GPU, VRAM.
- Latency target: an occasional personal chat tolerates slower token generation than an interactive application.
- Supporting services: reserve capacity for Docker, the reverse proxy, monitoring and the operating system.
- Storage growth: model files, chat attachments, logs and backups need more than the initial model download.
Start with a CPU VPS sized for a single-user test, measure response time and memory pressure, then scale vertically if context or concurrency causes swapping. A GPU is optional, not a prerequisite. Select a GPU plan only when measured CPU latency or concurrent demand warrants its recurring cost, and confirm that the provider supports the required NVIDIA container runtime if you choose CUDA.
Rank #2
- WINDEX GAMING WIPES: effortlessly clean all your gaming gear and gaming electronics in one easy step with our new Windex Gaming Wipes
- AMMONIA FREE: our ammonia-free formula is safe to use on all of your favorite gaming equipment like consoles, headsets, controllers, screens, keyboards, and more
- PRE-MOISTENED WIPES: each wipe instantly removes dust, dirt, and stubborn smudges and fingerprints leaving a streak-free shine
- ANTI-STATIC: great for cleaning and wiping down your gaming gear and electronics leaving behind no visible residue
- COMPACT RESEALABLE PACKAGE: store your pack of Windex Gaming Wipes anywhere in your home for whenever a quick and easy wipedown is needed
Prepare the Linux VPS and Docker network
- Use a current Linux distribution with Docker Engine and the Docker Compose plugin installed. Apply security updates and create a non-root administrative account.
- Allow SSH and the reverse proxy’s HTTPS port in the VPS firewall. Do not open TCP 11434 to the Internet for the normal layout.
- Create a deployment directory, for example
mkdir -p ~/phi3-stack && cd ~/phi3-stack. - Generate a long, stable Open WebUI secret:
openssl rand -hex 32. Store the result in a protected.envfile. ChangingWEBUI_SECRET_KEYlater can invalidate existing sessions.
If a reverse proxy runs on the same VPS, bind the WebUI port to loopback (for example, 127.0.0.1:3000:8080) instead of publishing it on every interface. The proxy then forwards HTTPS traffic locally.
Deploy Ollama and Open WebUI with Compose
Create .env in the deployment directory:
WEBUI_SECRET_KEY=replace-with-the-output-of-openssl-rand-hex-32Then create compose.yml:
services:
ollama:
image: ollama/ollama:latest
restart: unless-stopped
volumes:
- ollama:/root/.ollama
networks:
- ai
open-webui:
image: ghcr.io/open-webui/open-webui:main
restart: unless-stopped
depends_on:
- ollama
ports:
- "127.0.0.1:3000:8080"
environment:
OLLAMA_BASE_URL: http://ollama:11434
WEBUI_SECRET_KEY: ${WEBUI_SECRET_KEY}
volumes:
- open-webui:/app/backend/data
networks:
- ai
volumes:
ollama:
open-webui:
networks:
ai:The ollama service name becomes the hostname Open WebUI uses on the private Docker network. The Ollama volume stores model files under /root/.ollama. The WebUI volume at /app/backend/data stores chats, users and settings, so replacing a container does not erase them.
Start the stack with:
docker compose up -d
docker compose psCheck logs if either service is not running:
docker compose logs --tail=100 ollama
docker compose logs --tail=100 open-webuiInstall Phi-3 in Ollama
Run the model commands inside the Ollama container:
docker compose exec ollama ollama pull phi3
docker compose exec ollama ollama list
docker compose exec ollama ollama run phi3ollama pull phi3 downloads the model, ollama list confirms that the tag is present, and ollama run phi3 opens a direct terminal test. Exit the interactive test after confirming that the model answers. The official model name is phi3; do not substitute a similarly named tag without checking the current Ollama library.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →If Ollama is installed directly on the host instead of in Docker, use the host commands without docker compose exec. The model still has to be present before Open WebUI can list it.
Connect Open WebUI to Ollama
Container-to-container connection
With the compose file above, the connection URL is http://ollama:11434. Open WebUI receives it through OLLAMA_BASE_URL. Open the WebUI in a browser at the proxy hostname (or temporarily at http://127.0.0.1:3000 through an SSH tunnel), create the first administrator account and open Settings → Admin → Connections → Manage Ollama API Connections. Verify the URL and refresh the model list. phi3 should appear after the pull completes.
Rank #3
- Ergonomic & Adjustable Design: Fight "text neck" and improve your posture. Designed for comfort, it offers an adjustable height (7.1in-8.5in) (1.4in adjustable range) and angle (5°-85°) Please note: cell phone stand is only 9 inches tall—the length of a standard pencil
- Without blocking subtitles Perfect for office desk Tiktok Face Time Zoom video—this is the ultimate kitchen desk essential for phones! Perfect viewing experience while reducing neck back strain, A must-have for gamers streamers desk, and anyone who values both style and support!
- Sleek & Stable Design fits Otterbox Case: Say goodbye to bent cables, tangles, and tip-overs! LISEN’s stylish adjustable desk phone holder features a built-in charging hole and smart cable management, keeping your phone secure and your workspace clutter-free cool gadgets for men
- ANTI-SLIP DESIGN free hands: The under 10.00 items easy gadgets back and the bottom of this kitchen essentials for women are fully covered by anti-skid silicone, which LISEN stand up desk decor can provide maximum protection device from any scratches and slides
- Your Perfect Viewing Companion:If your product arrives damaged or breaks within 365 days, reach out to LISEN for a quick and hassle-free solution. Struggling with assembly, like tightening the screw? We’ve got your back! Your experience matters to us. (Product including: phone Holder stand*1, screw*1, hexagon screwdriver*1, specification*1)
Host Ollama with Dockerized WebUI
Use Docker’s host-gateway mapping and point the WebUI at the host:
docker run -d -p 3000:8080
--add-host=host.docker.internal:host-gateway
-v open-webui:/app/backend/data
-e OLLAMA_BASE_URL=http://host.docker.internal:11434
-e WEBUI_SECRET_KEY="$(openssl rand -hex 32)"
--name open-webui --restart always
ghcr.io/open-webui/open-webui:mainFor a host-installed Ollama daemon, configure its listener so Docker can reach it. Bind that listener to a private interface or enforce a firewall rule; a listener on 0.0.0.0:11434 without filtering is a public unauthenticated model API.
Recommended Free Tools
Ollama on another server
Set OLLAMA_BASE_URL to the reachable URL of the Ollama server and restrict that path with network controls and authentication appropriate to your design. Test connectivity from inside the WebUI container rather than from your laptop, because container DNS and firewall paths can differ.
Secure the public entry point
Put Open WebUI behind a reverse proxy that obtains a certificate, redirects HTTP to HTTPS and forwards WebSocket connections. Keep the proxy’s upstream pointed at the loopback WebUI binding or the internal Docker network. Add an authentication layer if the application will be reachable by more than a trusted private group.
Ollama’s port 11434 should remain on the Docker network in the separated layout. If an external client genuinely needs the API, expose it through a separate authenticated, rate-limited route rather than opening the raw port. Review firewall rules after every networking change.
CPU, CUDA and image variants
Open WebUI publishes CPU-capable, CUDA and Ollama-bundled image variants. CPU inference avoids GPU driver and runtime maintenance and is usually the least expensive option. CUDA can reduce latency for concurrent users, but it requires a compatible NVIDIA GPU, Docker GPU support and the NVIDIA container runtime/toolkit. A CUDA-capable WebUI image does not by itself provide a GPU; the VPS, host drivers and container configuration must all support it.
Rank #4
- Easy Installation: Our gaming lights have been designed to enhance your entertainment experience. Simply position the led light bar next to your TV or game console to create a game room environment with ease. They come with USB power and a 60 inch cord for direct mounting and mobility
- Audio Syncing & Extensive Mode: The built-in high-sensitivity music sensor allows you to sync the TV led lights to your favorite music. Lights react and move to the beat of music, film and game audio, creating an incredible immersive sound and light experience – and all without the need for an additional hub. You can choose from 64+ patterns, including strobe, illusion, jump, energy, gradient, running, chasing, fluttering mode etc. Each mode can be applied to different scenes
- 64+ Chasing IC+RGB & W Effect: The independent control design of IC+RGB and white all-in-one multifunctional rgbw led bar lights monitor backlight allows you to easily switch the led lights for desk between IC+RGB and white to enjoy a variety of lighting effects, you can choose white lighting mode with high brightness for better illumination when working, etc. Choose IC+RGB lighting mode when you need to create an entertaining atmosphere for the light show, 16 million colours of Color changing rgb light bar let you dance to the beat and create a dynamic party atmosphere
- Dynamic and Dimmable Lights: 1%-full brightness free brightness adjustment gives you the flexibility to adjust dimmable led light bar for room decorations or a variety of occasions
- USB Powered: USB light bars for room can be powered by powernank, usb charger, TV, Monitor, laptop PC mainframe, making behind TV led lighting bar easy to install in any location.TV led lights are powered by the TV usb interface / PC mainframe, can be turned on/off synchronously with TV / PC
Image tags include main, slim, cuda and ollama. Select deliberately. For reproducible upgrades, pin a tested release tag when the project’s release process provides one instead of silently tracking a moving tag.
Persistence, backups and upgrades
Back up both named volumes. Losing the WebUI volume removes chats, users and settings; losing the Ollama volume requires downloading the model again and may remove locally created model data.
docker run --rm
-v phi3-stack_open-webui:/data
-v "$PWD":/backup
alpine tar czf /backup/open-webui-backup.tgz -C /data .
docker run --rm
-v phi3-stack_ollama:/data
-v "$PWD":/backup
alpine tar czf /backup/ollama-backup.tgz -C /data .Volume names are project-dependent; check them with docker volume ls before running a backup command. Store archives off the VPS and periodically test restoration.
For an update, back up first, then run:
docker compose pull
docker compose up -dInspect logs and sign in after the update. Do not recreate the stack with a different volume declaration or anonymous storage, or the application can appear empty even though the old data still exists in another volume.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Troubleshoot the failures that look like missing models
“No model” in the selector
- Run
docker compose exec ollama ollama listand confirm the exactphi3tag. - Check
OLLAMA_BASE_URLin the WebUI container withdocker compose config. - Confirm the Ollama container is running and reachable on port 11434 from the WebUI network.
- Refresh the connection page at Settings → Admin → Connections → Manage Ollama API Connections.
Connection refused
Inspect whether Ollama is listening only on loopback. A Dockerized WebUI cannot reach a daemon bound exclusively to the host’s 127.0.0.1. Check container DNS, host-gateway configuration and firewall rules. In the separated compose layout, the hostname must be ollama, not localhost.
Chat hangs behind a proxy
Configure WebSocket forwarding and increase streaming and idle timeouts in the reverse proxy. A proxy that buffers or terminates long-lived connections can make a healthy model look frozen.
Best Value
- Compatible with Wide Screens - To ensure compatibility with the dual monitor mount, your each monitor must meet three conditions at the same time: First, computer screens size range: 13 to 32 inches. Second, screen weight range: 4.4 to 19.8 lbs. Third, the back of the monitor screen must have VESA mounting holes with a pitch of 75x75mm or 100x100mm.
- Regarding the compatibility with desks - Your desk must meet three conditions at the same time: First, desk material: Only wooden desks are recommended, plastic or glass desks cannot be used. Second, desk thickness range: 0.59" - 3.54". Third, the bottom of the desk should not have any cross beams or panels, as this will interfere with installation. We recommend carefully checking that your desk and monitors meets all above conditions before purchasing.
- Dual C-Clamp Hold - Worried your dual monitors might wobble or slip? Our upgraded base uses a larger platform plus a dual C-clamp structure to lock the dual monitor arm firmly to your desk. Each arm safely keeps your screens steady while you type, click and game—no shaking, no sliding, just a clean and secure setup you can trust every day. It also provides Grommet Mounting installation choice, both options ensure stable and secure fixation for your 0.59" - 3.54" desk.
- Full-Motion Adjustment For Comfortable View - Pull the screen closer when you’re deep in a spreadsheet, push it back to watch videos, or rotate to portrait for coding — moving everything smoothly with just one hand. The monitor stand offers +85°/-50° tilt, ±90° swivel and 360° rotation. Raise your monitor up to 15.75″ to support a healthy sitting posture. Whether you’re working from home, gaming through the night, or switching between video calls and documents, getting the screens to your natural line of sight helps relieve neck, shoulder and back strain so you can stay focused longer with less fatigue.
- Keep Your Desk Organized: By lifting both screens off the desktop, this dual monitor stand opens up valuable space for your keyboard, notebook, docking station or a simple, clutter-free work area. Built-in cable management guides wires along the arms, keeping cords out of sight and out of the way. Enjoy a tidy, modern workstation that looks as good as it feels to use.
Chats or users disappeared after an update
Verify that /app/backend/data is mounted from the intended named volume and that the replacement container uses the same project and volume name. Check docker volume ls for an accidentally created anonymous volume, then restore the latest known-good backup if necessary.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesResponses are slow or run out of memory
Reduce context length and simultaneous requests, watch host memory and CPU, and add swap cautiously. If the workload still exceeds the machine, move to a larger VPS or a GPU plan. Do not infer expected performance from the 3.8-billion-parameter count alone.
Operate the service safely
- Keep Docker, the operating system and the reverse proxy patched.
- Restrict administrator account creation and review active users in Open WebUI.
- Monitor disk usage for model files, logs and backups.
- Use a private network or VPN for administration when possible.
- Record the image tags, compose file and environment values needed to reproduce the deployment, while keeping the secret key out of public repositories.
- Test a restore on a separate volume before relying on backups during an incident.
When this deployment is the right fit
This stack is well suited to a private assistant, internal documentation search or a small team that wants browser access without sending prompts to a hosted service. CPU-only operation keeps the initial setup straightforward. A GPU becomes reasonable when several users generate at once or when measured CPU latency prevents interactive use.
A bundled image minimizes the number of commands for a disposable experiment. Separate services are the better long-term choice when you need independent upgrades, clear backups, or the option to move Ollama to another machine. In every layout, the essential invariants are the same: install phi3 in Ollama, point Open WebUI at the correct API URL, persist both data directories and protect the API from unsolicited Internet traffic.
Frequently Asked Questions
Can I run Phi-3 on a CPU-only VPS?
Yes. Phi-3 Mini does not require a GPU. CPU inference is simpler and often cheaper; choose a GPU only if measured latency or concurrent demand justifies the additional infrastructure and cost.
Why does Open WebUI show no models immediately after installation?
Open WebUI does not include models. Pull phi3 into Ollama, verify it with ollama list, and confirm that the WebUI connection uses the Ollama service hostname and port 11434.
Should port 11434 be opened in my VPS firewall?
Normally no. Keep it on the private Docker network and publish only the HTTPS reverse-proxy entry point. Expose Ollama externally only for a deliberate, authenticated remote-client architecture.
What must be backed up?
Back up the Ollama volume mounted at /root/.ollama and the Open WebUI volume mounted at /app/backend/data. The first contains models; the second contains chats, users and settings.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
