Open WebUI

What is Open WebUI?

Open WebUI is a self-hostable AI chat, knowledge base, and team collaboration platform; it is often used to provide a browser interface for local Ollama models, and it can also connect to services such as OpenAI, Anthropic, Gemini, Azure, vLLM, as well as other services that are compatible with OpenAI or Open Responses.

Users can manage multi-model conversations, document RAG, web search, voice, image generation, tools, and automated tasks through a unified interface.

Open WebUI does not include a proprietary large model that can directly replace all models. It acts as an application layer between the models and users; the quality, speed, context handling, and cost of the responses depend on the inference service being used.

By using local Ollama, the main data remains on your own device or server; whereas when cloud-based models or external tools are used, the relevant data is sent to the corresponding provider.

Core functions of Open WebUI

Unified chatting for multiple models

Administrators can access multiple local and cloud-based models, while users can switch between these models within the same session or compare the responses side by side. The chat feature supports file and image uploads, message queues, folders, tags, prioritization, memory functions, voice input and output, as well as a variety of other operations on content.

The model preset allows the integration of system prompts, knowledge bases, tools, and parameters, thereby creating specialized assistants for writing, teaching, programming, or customer service.

Ollama and local models

Open WebUI can be connected to Ollama, which is located on the host machine or a remote server; there are also integrated container images that include Ollama pre-installed. Local inference does not incur cloud API fees based on the number of tokens used, but it does consume CPU, GPU, memory, storage resources, and electricity.

The localhost within the container refers to the container itself; when connecting to the host machine Ollama, it is necessary to use the correct address of that host machine and to limit the scope of listening.

Knowledge base and RAG

Users can upload documents and create knowledge collections, allowing the model to retrieve relevant excerpts during conversations. The system supports vector retrieval, a combination of keyword and vector retrieval, reordering, insertion of full content, as well as various document extraction engines.

External knowledge sources can also be connected directly to Qdrant, Milvus, or pgvector, thereby avoiding the need to re-import existing indexes.

If the default value for the context size in Ollama models remains low, a large amount of retrieved content may be truncated, resulting in poor RAG performance. Deployers need to adjust the context size, segmentation, number of results to be retrieved, and reordering based on the capabilities of the model; they should also verify answers and references using standard questions, rather than simply checking whether the documents have been uploaded successfully.

Online searching and web page content

Open WebUI can be connected to search services such as SearXNG, Google, and Tavily, using the content of web pages as the context for the model. The accuracy, quantity, and privacy of the search results are determined by the respective search engines.

When encountering news, prices, policies, and professional information, users should still visit the original page to verify the timing and context.

MCP, OpenAPI, and tools

The platform can connect to HTTP MCP, access stdio MCP via a proxy, and also discover external interfaces using the OpenAPI specification. Workspace Tools are Python scripts that run directly within the Open WebUI process; they are very powerful, but this gives equivalent capabilities to those of server-side code execution.

Only trusted administrators can create or import them.

The fact that a community tool is marked as popular or selected does not mean it has undergone security audits. Before installing it, one should read its code, check the dependencies and key usage, and restrict the permissions of the Workspace.

For tasks that require stronger isolation or independent scaling, it is preferable to run the tool as an external MCP or OpenAPI service.

Functions and extended systems

The Pipe Function allows for the addition of model providers or custom request logic; Filters can be used to perform tasks such as data masking, formatting, rate limiting, and log processing before and after messages are processed; Actions enable the addition of buttons related to messages. Older Pipeline projects are still compatible with existing deployments, but they have been marked as obsolete by the developers – new projects should prefer to use Functions, Tools, or external tool servers.

Code, images, voice, and automation

Open WebUI can be connected to browser-based Python environments, isolated Open Terminal sessions, image generation services, as well as speech recognition and synthesis tools. It also supports scheduled execution of prompts, structured task lists, and message notifications.

When code is executed or terminal tools have access to real files and the network, isolated containers, resource restrictions, and per-user boundaries should be used; browser sandboxes cannot be regarded as a complete security barrier.

Team Management and APIs

Administrators can manage users, groups, model access permissions, announcements, usage analytics, blind testing, and Webhooks. The platform provides API keys with user-specific permissions, as well as interfaces for chatting, file handling, accessing knowledge bases, and invoking various tools.

The key inherits the permissions of its creator; it is necessary to restrict the accessible endpoints, establish an expiration policy, and carry out regular rotations.

Comparison of Open WebUI versions and costs

Standard self-hosting comes without any software subscription fees, but servers, models, and third-party services are not necessarily free. For enterprise white labeling, brand customization, exclusive features, or formal business partnerships, licenses are required; no fixed prices are available officially, and quotes are provided based on the organization and number of seats.

Comparison of model integration methods

Open WebUI license details

The current official license is based on BSD 3-Clause, with additional requirements regarding brand protection. It can be used free of charge for internal deployment, provided that the original brand of Open WebUI is maintained intact.

If no more than 50 users are active within any consecutive 30-day period, it is possible to adjust the brand in accordance with the official guidelines; those who make substantial contributions can also obtain the right to use a specific brand, either by receiving written permission or by purchasing an enterprise license.

For white-labeling, enhanced branding that goes beyond the free exception limits, displaying one’s own logo alongside the official brand, or using proprietary enterprise features, a corporate license is required. The license may evolve as project versions change; before distributing modified versions, deploying them for clients, or using them on a large scale, it is necessary to check the LICENSE included with the version in use as well as the official guidelines.

Open WebUI Docker Deployment Tutorial

1. Installing Docker:Install the official Docker environment on Windows, macOS, or Linux to ensure that you can pull official images.
2. Select image:The standard image is suitable for most users; the slim image has a smaller size, the CUDA image supports NVIDIA GPUs, and the Ollama image includes Ollama.
3. Create a persistent volume:Map the data directory to the backend data path within the container, thereby preventing the loss of databases, chat records, and knowledge bases when the container is rebuilt.
4. Start up and create an administrator:The account registered on port 3000 of this device becomes the administrator when it is opened for the first time. Do not leave the page corresponding to the first registration unattended on a public network.
5. Connection model:Configure the Ollama address or the Base URL and key for cloud models, and first test normal conversations as well as streaming output.
6. Configure RAG:Select document extraction, Embedding, and vector databases, and use a small amount of Chinese data to assess the quality of parsing and retrieval.
7. Reinforcement services:Disable unnecessary registrations, and configure HTTPS, strong encryption keys, a reverse proxy, a firewall, access control, and backups.
8. Fixed production version:main and latest are rolling images; in a production environment, a specific version label should be used. Back up the data before upgrading and read the migration instructions.

Open WebUI Usage Guide

Create reusable professional workflows

1. Break down complex topics into four categories of questions: background, data, comparison, and conclusions;
2. A combination of multiple models for unified chatting, along with Ollama, local models, knowledge bases, and RAG, constitutes the fixed research steps;
3. Give priority to using the official website, research papers, regulatory documents, and raw data;
4. A second person is assigned to review conclusions that are considered high-risk;
5. Save queries, evidence, versions, and unresolved issues;
6. Re-run after the data changes and update the conclusions;

Precautions for large-scale deployment

• A single instance can use the default database and local storage;
• Multi-instance production deployment requires PostgreSQL, Redis, a concurrently accessible vector database, and shared file storage.
• Each instance must use consistent Secret and connection configurations, and database migration can only be carried out by one instance.
• Model throttling, audit logs, resource quotas, backup and recovery procedures, as well as upgrade rollback processes should also be established.

Which users are it suitable for

• Individuals who wish to add a full web chat interface for Ollama or other local models;
• AI teams that need to integrate with multiple cloud-based and self-hosted models in a unified manner;
• Organizations that plan to establish internal portals for document querying, searching, and working with models;
• Developers who need MCP, OpenAPI, Python tools, and message filtering extensions;
• A team for private deployment that is capable of managing identity, infrastructure, and data security.

Advantages and usage limitations

• The advantages of OpenWebUI are its mature user interface, wide range of model compatibility, strong capability for privatization, as well as integration of a knowledge base, various tools, search functions, voice capabilities, image processing tools, and team management features.
• It is suitable for a smooth expansion from personal local chats to internal organizational platforms;
• The limitations lie in the large number of functions and extensive configuration options, which result in a high level of complexity in production and operation.
• Plugins can execute any code, and insufficient permissions may put the server at risk;
• Self-hosting does not automatically mean security or compliance;
• Model responses and RAG citations may still be incorrect, and critical information must be verified manually;
• Before upgrading, back up your data and prioritize installing a stable version that includes security fixes;

Frequently Asked Questions

Is Open WebUI free?

Standard self-hosting and internal use while retaining the original brand are free of charge. Servers, GPUs, cloud model APIs, and third-party services are charged separately;

For white-label solutions, strong brand presence, and enterprise-specific capabilities, it is necessary to request a quote in order to obtain an enterprise license.

Is it necessary to install Ollama in order to use Open WebUI?

It’s not necessary. It can connect to Ollama, as well as directly to OpenAI-compatible services, Open Responses, and various cloud-based model services.

Does Open WebUI support MCP?

HTTP MCP is supported; stdio MCP can also be accessed via a proxy. The code and permissions related to external services require separate auditing.

Can Open WebUI operate completely offline?

It is possible, but local models, local embeddings, and local vector databases must be used, while online searches and all external APIs must be disabled. The process of installing mirrors and downloading models usually still requires an internet connection.

Is Open WebUI suitable for deployment on the public internet?

It is possible, but it is necessary to configure HTTPS, authentication, strong secrets, a reverse proxy, a firewall, backup mechanisms, and least-privilege principles. Do not expose default ports or plugins with high permissions to the public internet.