Open WebUI is a front end. It does not run models: its docs describe pointing it at Ollama or an OpenAI-compatible endpoint, and Docker is the officially supported and recommended way to run it, with Python and Kubernetes documented as alternatives.
That layering is the whole shape of the migration. If you try to "move Open WebUI" as one thing you will get confused, because the two halves live in completely different places and only one of them is actually at risk.
What does Open WebUI actually hold?
Your conversations, prompts, knowledge-base documents and their embeddings, and your user accounts. All of it persists in a single Docker volume named open-webui, mounted inside the container at /app/backend/data, per the quick-start command in Open WebUI's own documentation. Delete the container without dealing with that volume and you have not lost anything yet. Delete the volume and you have.
What it does not hold is the models. Those belong to the backend. If you were running the usual Mac setup, they are in Ollama's store, and that is a completely separate migration with its own guide.
The macOS problem nobody mentions
On a Mac, Docker runs inside a Linux virtual machine. So the path that docker volume inspect reports as the volume's mountpoint is a path inside that VM, not a folder on your Mac, and you cannot open it in Finder or cd to it in Terminal. This is the step where most Mac migrations stall, and it is not an Open WebUI issue.
The way across the boundary is to have Docker do the copying for you, either out of the running container or out of the volume directly. Both are one command and both are in the steps below. What does not work, and what people try first, is navigating to the mountpoint path. It does not exist on your side of the VM.
Do the reversible half first
Before any container work, use Open WebUI's own interface while it is still running. Export the conversations you care about, and pull down any documents you uploaded to a knowledge base if you no longer have the originals elsewhere.
This matters more than the volume backup, because an application-level export gives you readable files. A volume archive gives you a database in a layout that is Open WebUI's, is version-specific, and is only useful for restoring into Open WebUI. Take both. Use the readable one.
The models are a different job
Whatever Open WebUI was pointed at still has your models after Open WebUI is gone, untouched. If that was Ollama, the weights are sitting in a content-addressed store as sha256-named blobs, which is its own recovery problem with its own walkthrough. Do not start it until the Open WebUI data is safely off the volume, because the two jobs share nothing and interleaving them is how you lose track of what you have finished.
The order that works: export from the UI, archive the volume, then deal with the backend, then tear down the container and the volume.
Getting off the stack entirely
Plenty of people running this setup on a Mac are running three things to get one thing: Docker for the container, Ollama for the inference, and Open WebUI for the window. That stack makes complete sense on a home server with several users, which is what Open WebUI is built for. On one Mac used by one person it is a lot of moving parts for a chat window.
If that is the situation, the replacement is a single native app rather than a smaller stack, and the containers and the second process both go away. The native options are worth reading before you pick one, because more than one of them is reasonable.
Landing in ToolPiper
ToolPiper is a native Mac app that bundles the runner, so there is no container and no separate model server. The engine is upstream llama-server embedded directly, build b10068, unmodified, and models are plain named GGUF files in ~/Library/Application Support/ToolPiper/models/. Nothing sits in a volume you need a command to read.
On the data question that this whole page is about: your chats live in the app on your Mac, not in a database inside a VM. And if you are keeping Ollama around during the transition, ToolPiper will connect to it as a backend so both libraries are usable from one interface while you decide.
The free tier is the whole runner: unlimited GGUF downloads, multi-model switching, a local OpenAI-compatible API, embeddings, developer tokens, all speech including transcription, text-to-speech, voice cloning and push-to-talk dictation, browser automation, and an MCP server with over 420 tools. Local RAG over your files is the $10 Pro tier, which is the closest equivalent to Open WebUI's knowledge bases and the one thing here that is not free.
Where Open WebUI stays the better choice, plainly: several people sharing one instance, a machine that is not a Mac, or a setup you reach from other devices on your network. It is genuinely open source, its local-by-default posture is real, and none of that is true of us. ToolPiper is macOS 26 or newer on Apple Silicon, one machine, one person.
A note on cost, because this is the one axis where the comparison is not about money. Open WebUI charges nothing and neither do we for the runner. What differs is what you have to keep running. The backend underneath Open WebUI is usually Ollama, and that stack does now have a free versus paid line, because Ollama Cloud is a priced product sitting beside the free local one. Where that line falls.