Open WebUI
A chat workspace that connects local or compatible API models and organizes conversations, model settings, knowledge bases and tools; desktop and server deployments have different boundaries.
What Open WebUI provides
Open WebUI wraps local models or compatible APIs in a chat workspace with conversations, model switching, knowledge bases and tool entry points. The archived desktop package and a server deployment are separate paths: a personal desktop test has a different account, container, database, proxy and storage boundary from a multi-user installation.
Data and permission boundaries
Chat, embeddings, vector storage, attachments, models and tools may each read data or use the network. A cloud model can receive prompts or document chunks, while custom functions may access files or make external requests. Back up data volumes and configuration before upgrades, and treat administrator, registration and reverse-proxy settings as security controls rather than visual options.
Maintenance note
This page reviews Open WebUI Desktop 0.0.20, Docker deployment, local model connections, knowledge bases, multi-user permissions, tools and backup boundaries. Content review date: 2026-08-23.
Save to your cloud drive
Open the cloud drive to get the file directly, or save it for convenient access on another device.
Quark Cloud Drive
RecommendedSave Open WebUI to this cloud drive
Baidu Netdisk
Save Open WebUI to this cloud drive
Open WebUI Desktop 0.0.20 and local model connection guide
Start in a non-sensitive test environment, complete one local model conversation and document retrieval test, then consider Docker, multi-user access or external tools.
Before you start
- The archived desktop package targets macOS Apple Silicon; other systems should use a controlled server deployment with a defined proxy.
- Prepare a local Ollama or compatible API, record the model name and access scope, and use limited test credentials for online services.
- Prepare public or fictional test files and identify chat, embedding, vector, attachment and log storage paths.
Installation steps
- 01
Install the desktop package or pin the server image
Install the macOS Apple Silicon DMG and verify version 0.0.20, or pin the server container and data volume before first start.
- 02
Connect a model service
Add local Ollama or a compatible API, run one minimal conversation and check the address, model name and authentication fields before saving.
- 03
Create least-privilege accounts
In a multi-user deployment, configure administrator, ordinary-user and invitation rules, and keep registration, external tools and sharing off until isolation is verified.
Quick start
- 01
Compare local and cloud paths
Ask the same harmless test questions against a local and a cloud model, and record the different network, latency and data-retention behavior.
- 02
Import one test document
Check parsing, embedding, citations and deletion with a public file; successful retrieval does not make a generated answer authoritative.
- 03
Back up volumes and configuration
Record container, database, model and storage-volume relationships, export settings and test a restore before replacing an existing deployment.
Usage tips
- The workspace, model service, embedding provider, vector database and tools can be separate components with separate data paths.
- Registration, reverse-proxy and administrator settings are deployment controls; hiding a button in the interface is not access control.
- Custom models, functions and tools may read files or call networks, so keep a source inventory and restrict their scope.
Troubleshooting and uninstall
Why does the page open but the model not respond?
Confirm the model service works locally, then check the WebUI address, model name, container network and proxy with a minimal request.
Why are knowledge-base citations inaccurate?
Use a short document to test parsing, chunking, embeddings, vector storage and cited passages, and keep the original file and index backup while tuning.
- Export data and revoke credentialsExport required conversations, knowledge-base lists, model settings and administrator configuration, then revoke external API keys and tool tokens.
- Stop the service and handle volumesQuit the desktop app or stop the server containers first, then handle databases, uploads, caches and model directories according to the backup plan.
Frequently asked questions
Is the Open WebUI desktop package the same as its Docker server?
No. The archived desktop package targets macOS Apple Silicon, while Windows, Linux and multi-user scenarios commonly use a server deployment with separate containers and data volumes.
Can Open WebUI connect to Ollama?
Yes. Configure the Ollama service address and model, but check binding, firewall and proxy scope before allowing other devices to access an unprotected API.
Does a knowledge-base upload always stay local?
It depends on the model, embedding, vector store and tool configuration. Cloud providers may receive prompts or document chunks, so test with non-sensitive files first.