Open WebUI v0.12.4 Released with Context Compaction & On-Demand Skills

Open WebUI v0.12.4 Released with Context Compaction & On-Demand Skills

At a Glance

Item Value
Repository open-webui/open-webui
Version v0.12.4
Published 2026-10-25
License NOASSERTION
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

Open WebUI, the web user interface for LLMs, has released its latest version “v0.12.4". Open WebUI is a Python-based web UI tool that supports various backends such as Ollama and OpenAI-compatible APIs, allowing users to operate LLMs intuitively in local environments.

This update introduces on-demand skill loading and improvements to context compaction. Notably, context compaction now works even in short interactions containing large amounts of tool output, resolving issues where conversations were interrupted by context length limit errors. This is a highly practical update for users who want to minimize prompt consumption or heavily utilize tool calling.

Key Changes

Support for On-Demand Skill Loading

A feature has been added where instruction prompts for skills selected in the chat or tied to models are loaded only when needed by the model, rather than being sent in full with every message submission. Skills directly mentioned within messages are loaded immediately as before. This feature is disabled by default and requires the built-in Skills tool to be used. This affects administrators who want to prevent context (token count) from being overwhelmed by skill instructions. To use it, enable “Load selected skills on demand" from “Admin Settings" > “General" in the administrator settings after updating, or set the environment variable “ENABLE_SKILL_LAZY_LOADING".

Context Compaction Visualization and Error Avoidance

It is now clearly possible to grasp the status of context compaction on the chat screen. A “Context compacted" display has been added at the position where older messages were summarized, and expanding it allows users to check the summarized content. Additionally, compaction progress and errors are displayed directly inside the chat rather than as popups.

In addition, the scope of compaction processing has been expanded. Previously, short conversations that did not meet the retention count of the “Retained Messages" setting (such as cases where a single question returned a massive amount of tool output) or temporary chats would skip compaction, leading to context length errors on subsequent messages or regeneration. Starting from this version, appropriate splitting and compaction are executed at the first follow-up message, preventing interruptions due to errors. This brings significant benefits to users running models that frequently utilize tools. No settings changes or reconversion work are required.

Administrator UI Item Hiding and Automatic Concealment of Invalid Items

A feature has been implemented allowing administrators to hide individual UI elements from regular users. This can be configured from “Admin Settings" > “Interface" under “Visible controls" or via the environment variable “UI_VISIBILITY". A wide range of items can be controlled, including model features, built-in tools, various attachment features in the “+" menu, commands like /status or /settings, usage information, and sidebar pin buttons. Items set to hidden retain their internal values, and the features themselves continue to function.

Furthermore, invalid options that cannot be executed in the current state—such as image capture buttons when selecting a non-Vision model, or note attachment features for unauthorized users—are now automatically hidden from the screen. This prevents UI confusion where clicking an unavailable button causes selected content to disappear. Administrators sharing instances with multiple people should check their display control settings as needed.

Improved Knowledge Base Upload Status Display and Management

Status displays when adding files to the knowledge base have been improved. Immediately upon reaching the server, the display switches from “Uploading" to “Processing", and the status updates automatically without requiring a screen reload even during processing such as text extraction. In addition, when multiple files are added simultaneously, they now immediately appear in the list as uploading. Furthermore, “Settings" > “Data Controls" > “Manage Files" now supports bulk deletion of multiple files and bulk downloading in ZIP format. This improves workflow efficiency for users who register local documents in the knowledge base and utilize RAG.

Support for the Latest MCP Protocol and Improved Usability

For Model Context Protocol (MCP) connections, the stateless “2026-07-28" protocol version is now supported. Compatibility with older MCP servers that use traditional handshakes is also maintained. Additionally, a wide variety of usability adjustments for daily use have been incorporated, such as the ability to select multiple conversations at once by holding Shift and clicking chats in the sidebar to archive or delete them collectively, support for swipe opening and closing of chat operation panels on mobile devices, and improvements to the text comparison screen ensuring left and right pane scroll positions do not drift.

How to Get It

Update to the latest version using either a Python environment (pip) or a Docker container, depending on your Open WebUI deployment environment.

Update Steps for Python Environment (pip)

If you are running directly in a Python environment, use the pip command to install or update. Python 3.12 is recommended to avoid compatibility issues.

Run the following command in your terminal to install or update the package:

pip install open-webui

After installation is complete, start the server by running the following command:

open-webui serve

Once the server starts up, you can access it from your browser at http://localhost:8080.

Update Steps for Docker Environment

If you are using Docker, specify the image suited for your use case and environment to start the container. To prevent database loss, it is mandatory to properly mount the volume by specifying the -v open-webui:/app/backend/data option.

If you are running Ollama on your host terminal, start the container with the following command:

docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main

To connect to Ollama running on a separate server, start it by specifying the target server’s URL in the OLLAMA_BASE_URL environment variable:

docker run -d -p 3000:8080 -e OLLAMA_BASE_URL=https://example.com -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main

To utilize CUDA acceleration via Nvidia GPU, install the Nvidia CUDA container toolkit on a Linux or WSL environment, and then use the official image with the :cuda tag:

docker run -d -p 3000:8080 --gpus all --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:cuda

If you only use the OpenAI API, start it by specifying the OPENAI_API_KEY environment variable:

docker run -d -p 3000:8080 -e OPENAI_API_KEY=your_secret_key -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main

To use the bundled image combining Open WebUI and Ollama, use the image with the :ollama tag. The command when utilizing a GPU is as follows:

docker run -d -p 3000:8080 --gpus=all -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:ollama

To run the bundled image with CPU only, execute the following command:

docker run -d -p 3000:8080 -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:ollama

After starting with the Docker command, you can access the operation screen by visiting http://localhost:3000 in your browser. Detailed instructions for configurations and updates using Docker Compose, Kustomize, Helm, etc., are provided in the Open WebUI Documentation.

Related Articles

What to Read Next

Sources