This page covers three of the routes in Open WebUI's Quick Start for a Mac. The Desktop app installs Open WebUI for you and has a macOS download for Apple silicon and one for Intel. Docker is what the docs call officially supported and recommended for most users. Python is what they call suitable for low-resource or manual setups. Open WebUI has no models of its own, so you connect a provider, and its docs say Ollama running on the same machine is picked up automatically, at host.docker.internal from Docker or localhost from Python. Two traps are named in those docs. Switching on the num_ctx control pre-fills 2048 tokens and overrides Ollama's own context length. And a Docker install without a fixed WEBUI_SECRET_KEY logs everyone out each time the container is recreated. This page repeats Open WebUI's and Ollama's documents and ran nothing on a Mac.
| Desktop app | Docker | Python (pip or uv) | |
|---|---|---|---|
| What the docs call it | Runs Open WebUI natively without Docker or manual setup | Officially supported and recommended for most users | Suitable for low-resource environments or manual setups |
| Install | Download the Apple silicon or Intel DMG from the Desktop repository, open the app and click Get Started | One docker run command, below | pip install open-webui then open-webui serve, or uvx with uv |
| Opens at | http://localhost:8080 | http://localhost:3000 | http://localhost:8080 |
| Needs | macOS 12 or later and 5 GB or more of disk; internet for installation and updates | Docker installed | Python 3.11 or 3.12; 3.13 is not supported |
| Your data | Not described in the Desktop README | In the open-webui volume | In the folder you name with DATA_DIR |
| Updates | Switch on Automatic desktop updates in Settings, About, to update on quit | Watchtower, or remove the container, pull the image and run it again | pip install -U open-webui, then restart |
The Quick Start's Desktop tab adds that for production deployments you install with Docker or Python. The Quick Start lists more tabs than this page covers: Docker Compose, Podman, Quadlets, Kube Play, Swarm, WSL, a Docker Desktop extension, Kubernetes with Helm, Conda, Venv and a third-party Pinokio.computer install. The Docker Desktop extension page says it is not an officially supported installation method.
The Quick Start's Docker tab says to replace your-secret-key with the output of openssl rand -hex 32, then run this command, which pulls the image and starts it:
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data -e WEBUI_SECRET_KEY=your-secret-key --name open-webui --restart always ghcr.io/open-webui/open-webui:main
| Flag | What Open WebUI says it does |
|---|---|
-p 3000:8080 | The UI is on port 3000 of your machine; change the left number if 3000 is taken |
-v open-webui:/app/backend/data | Your chats, users and settings live in this volume; it survives updates, and the page says never to run without it |
--add-host=host.docker.internal:host-gateway | Lets the container reach Ollama, or anything else, running on the host machine |
--restart always | Comes back after a reboot |
-e WEBUI_SECRET_KEY=your-secret-key | Set it once and keep it; without a fixed key every recreated container logs everyone out |
To update, the Quick Start says to pull the new image and recreate the container with the same command and the same WEBUI_SECRET_KEY, and that your data is in the volume. It also says to back up the volume before every update.
Ollama's FAQ says GPU acceleration is not available for Docker Desktop in macOS, because of the lack of GPU passthrough and emulation. That sentence is in the FAQ entry about GPU acceleration for Ollama's Docker container. Open WebUI's docs also offer an :ollama image that bundles Ollama inside the container, and this page does not say what that image does on a Mac.
pip install open-webui
open-webui serve
http://localhost:8080, and that DATA_DIR chooses where your data lives, for example DATA_DIR=~/.open-webui open-webui serve..webui_secret_key in the directory you start it from. Set WEBUI_SECRET_KEY, or always start from the same directory; a new key signs everyone out.DATA_DIR, the data is inside the environment, in site-packages/open_webui/data, and pip uninstall leaves it there.With uv, the docs give this command for macOS and Linux. Their tip says to set DATA_DIR, because without it uvx might store your data in a temporary folder that is deleted when the process ends.
DATA_DIR=~/.open-webui uvx --python 3.12 open-webui@latest serve
The same page installs uv itself with curl -LsSf https://astral.sh/uv/install.sh | sh, which runs a downloaded script. Open WebUI's page carries no warning for that line.
http://host.docker.internal:11434 from Docker, or http://localhost:11434 from Python, and shows under Manage Ollama API Connections. It does not say what the Desktop app picks.-e OLLAMA_BASE_URL=http://server:11434, or you can set the URL on the connection.The same page has a tip for an empty model list: Ollama on the host has to listen on 0.0.0.0 when Open WebUI runs in Docker. Its troubleshooting page says Ollama is likely only listening on 127.0.0.1, and that setting OLLAMA_HOST=0.0.0.0 makes it listen on all network interfaces. Ollama's FAQ says the server binds 127.0.0.1 port 11434 by default, and that when Ollama runs as a macOS application you set a variable with launchctl setenv and then restart the Ollama application:
launchctl setenv OLLAMA_HOST "0.0.0.0:11434"
That is Ollama's own example line. Neither Open WebUI's nor Ollama's page says what listening on all interfaces exposes. The Quick Start's sentence that Ollama is picked up automatically and its tip that Ollama must listen on 0.0.0.0 are both on the page, and this page did not test which applies to Docker on a Mac. For a container that cannot reach Ollama, the troubleshooting page also gives docker run -d --network=host with OLLAMA_BASE_URL=http://127.0.0.1:11434, after which it says the UI is at http://localhost:8080, and gives no note about Docker on a Mac.
Open WebUI's Ollama page says the num_ctx setting is unset by default, so Ollama's OLLAMA_CONTEXT_LENGTH applies. If num_ctx is set anywhere in Open WebUI, it sends it on every request and it overrides the server's value, and switching the control on pre-fills 2048, which the page calls very small. A context that is too small silently truncates the prompt, and with native function calling the model can return a blank response or a raw tool-call blob. To set a default for all Ollama models, the page says to set OLLAMA_CONTEXT_LENGTH on the Ollama server and leave num_ctx unset. Ollama's own context defaults are on Ollama context length on a Mac.
pip install -U open-webui, then open-webui serve again. If you run with UVICORN_WORKERS above 1, the docs say to start the new version with UVICORN_WORKERS=1 and wait until the migrations finish.docker rm -f open-webui removes the container. docker volume rm open-webui removes the volume, and the page marks that step as deleting all data.pip uninstall open-webui removes the package; the page lists deleting the DATA_DIR folder as an optional step that deletes all data.Open WebUI's app page says every instance is a Progressive Web App. For a desktop it lists installing from Chrome or Edge; its Safari steps are for iPhone and iPad, and it does not list Safari on a Mac. The Quick Start says a stock install makes three outbound calls on its own: a version check to GitHub, switched off with ENABLE_VERSION_UPDATE_CHECK=false; a model-list request to the default OpenAI connection, gone once you delete that connection or set ENABLE_OPENAI_API=false; and an update check for the local embedding model on Hugging Face at every start, switched off with RAG_EMBEDDING_MODEL_AUTO_UPDATE=false. The Desktop README says Desktop does not collect usage data and that an external AI provider's privacy policy applies if you connect one.
| Claim | Source |
|---|---|
| The three routes, the ports, the Ollama connection, the tips and outbound calls | Open WebUI docs, Quick Start |
| The docker run command, its flags and the Docker uninstall | Open WebUI docs, Quick Start (Docker tab) |
| Docker updating and the WEBUI_SECRET_KEY warning | Open WebUI docs, Quick Start (Docker, updating) |
| Python versions, pip, uv, data folder, secret key, updating and uninstall | Open WebUI docs, Quick Start (Python) |
| Ollama connection steps, the num_ctx override and its 2048 pre-fill | Open WebUI docs, Ollama |
OLLAMA_HOST, host.docker.internal and --network=host | Open WebUI docs, Connection errors |
| Desktop app downloads, requirements, updates and privacy | Open WebUI Desktop README |
| Installing an instance as a Progressive Web App | Open WebUI docs, Open WebUI as an App |
| The README's install lines and licence pointer | Open WebUI README |
| Ollama's default bind address, the macOS variable line and GPU in Docker Desktop | Ollama docs, Local FAQ |
0.0.0.0 when Open WebUI runs in Docker, in a tip for an empty model list. Ollama's FAQ gives launchctl setenv OLLAMA_HOST for the Mac app, followed by a restart of the Ollama application. This page did not test it.num_ctx pre-fills 2048, overrides Ollama's context length, and that a context too small silently truncates the prompt.