Overview

Relevant source files

• CHANGELOG.md
• LICENSE
• README.md
• TROUBLESHOOTING.md
• package-lock.json
• package.json
• scripts/prepare-pyodide.js
• src/app.css
• src/lib/apis/images/index.ts
• src/lib/components/chat/SettingsModal.svelte
• src/lib/components/common/RichTextInput.svelte
• src/lib/stores/index.ts
• src/routes/(app)/+layout.svelte/+layout.svelte)
• src/routes/(app)/+page.svelte/+page.svelte)
• src/routes/(app)/c/[id]/+page.svelte
• src/routes/(app)/playground/images/+page.svelte/playground/images/+page.svelte)
• src/routes/+layout.svelte
• static/manifest.json
• static/pyodide/pyodide-lock.json

Purpose and Scope

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform designed to operate entirely offline while providing a unified interface for multiple LLM runners README.md1-15 It serves as a comprehensive AI deployment solution, integrating local inference engines like Ollama and OpenAI-compatible APIs (including LMStudio, GroqCloud, Mistral, and OpenRouter) README.md15-33

The platform is built to handle complex AI workflows, including Retrieval Augmented Generation (RAG), native Python function calling, and real-time collaborative features, making it suitable for both individual power users and enterprise environments README.md46-67

---

Core Capabilities

Open WebUI distinguishes itself through a broad feature set that bridges the gap between raw model inference and a polished user experience.

AI Interface and Model Management

• Redesigned Interface: Visually rebuilt from the ground up in v0.11.0 with narrower conversation columns and improved typography CHANGELOG.md12
• Multi-Model Conversations: Effortlessly engage with various models simultaneously to compare strengths and responses README.md68-69
• Sub-Agents: Background helper agents that run their own tool-driven conversations and report results back to the main chat CHANGELOG.md13
• Model Customization: Wrap base models with custom instructions, tools, and knowledge to build specialized agents README.md37-38

Retrieval Augmented Generation (RAG)

• Vector Database Support: Integration with 9 vector database options including ChromaDB, PGVector, Qdrant, and Milvus README.md74-75
• Local RAG Integration: Supports hybrid search (BM25 + vector) with reranking and full-context mode README.md60-61
• Web Search Integration: Injects live results from dozens of providers (SearXNG, Brave, Google PSE, Tavily, etc.) into the chat experience README.md62-63
• Persistent Memory: The AI remembers facts about the user across conversations README.md44-45

Advanced Tools and Extensibility

• Plugin Primitives: Extensibility via Filters, Actions, Pipes, Tools, and Skills README.md36-37
• Scheduled Automations: Recurring AI tasks scheduled using iCalendar RRULE syntax, surfaced in the Calendar workspace README.md50-51
• Image Generation: Integrated support for OpenAI DALL·E, ComfyUI, and AUTOMATIC1111 README.md66-67
• Code Execution: Code block execution system that allows running Python code directly from chat messages via Pyodide package.json138 src/lib/stores/index.ts100

Enterprise and Collaboration

• Channels: Real-time shared spaces for team and AI collaboration with threads, reactions, and pins README.md42-43
• Notes: Dedicated workspace with a rich editor (TipTap) and AI-powered rewriting README.md40-41
• Granular RBAC: Administrators define detailed roles, groups, and permissions README.md34-35
• Authentication: Full support for LDAP/Active Directory, SCIM 2.0 provisioning, and SSO via OAuth or trusted headers README.md76-77

Sources: README.md1-85 CHANGELOG.md12-16 package.json54-162 src/lib/stores/index.ts100

---

System Architecture

Open WebUI follows a modern full-stack architecture designed for high responsiveness and horizontal scalability.

High-Level Component Interaction

The following diagram maps the logical components to their specific code entities.

Sources: src/routes/+layout.svelte165-174 src/lib/stores/index.ts13-41 src/lib/components/chat/SettingsModal.svelte8-19 README.md72-83 src/lib/components/common/RichTextInput.svelte130-153

---

Technology Stack

Core Technologies by Layer

Layer	Technology	Key Code Entities
Frontend	Svelte 5 / Tailwind CSS 4	package.json:44, 47, src/app.css:1
Real-time	Socket.IO	socket store in src/lib/stores/index.ts:31, socket.io-client in package.json:140
Editor	TipTap / Prosemirror	RichTextInput.svelte, package.json:67-85
Markdown	Marked / Turndown	package.json:121, 145, src/lib/components/common/RichTextInput.svelte:2, 32
Execution	Pyodide (Wasm Python)	package.json:138, pyodideWorker store src/lib/stores/index.ts100
Persistence	SQLite / PostgreSQL	README.md:72-73

Code Entity Space: Data Flow and Connectivity

The frontend communicates with the backend via a reverse proxy design to handle CORS and security.

Sources: src/routes/+layout.svelte165-174 src/lib/stores/index.ts31 README.md32-33 package.json140

---

User Configuration and Personalization

Users manage their experience through a comprehensive settings system:

• Interface Customization: Controls for UI scale (--app-text-scale) and responsive typography src/app.css16-24
• Rich Text Preferences: Custom rules for converting TipTap HTML back to Markdown using TurndownService, including specific rules for task lists, tables, and mentions src/lib/components/common/RichTextInput.svelte45-119
• Global State Management: Svelte stores manage everything from user sessions and settings to activeUserIds and chatId src/lib/stores/index.ts13-130
• Notification Targets: Webhook destinations for chat events, channel messages, and calendar alerts CHANGELOG.md16
• Application Lifecycle: Handled in the root layout, including socket initialization, version checking, and i18n setup src/routes/+layout.svelte165-234

Sources: src/app.css16-24 src/lib/components/common/RichTextInput.svelte45-119 src/lib/stores/index.ts13-130 src/routes/+layout.svelte165-234 CHANGELOG.md16