Open WebUI is a self‑hosted, extensible AI platform designed to run entirely offline while providing a provider‑agnostic interface for both local Ollama models and any OpenAI‑compatible API (e.g., LMStudio, GroqCloud, Mistral, OpenRouter, vLLM). Installation can be performed via pip, uv, Docker, Kubernetes (using kubectl, kustomize, or Helm), with official images tagged :ollama for CPU‑only setups and :cuda for GPU‑accelerated deployments requiring the NVIDIA Container Toolkit. The system offers granular role‑based access control, plugin architecture (Filters, Actions, Pipes, Tools, Skills) that integrates external services through MCP, MCPO, or OpenAPI tool servers, and enables the creation of specialized agents by wrapping base models with custom instructions, tools, and knowledge bases. Agentic execution is supported through an integrated terminal and isolated filesystem, scalable to multi‑user environments with per‑user resource limits via Terminals (Enterprise). Additional capabilities include persistent memory across chats, real‑time workflow visualization, calendar‑driven AI scheduling, automated prompt runs, responsive PWA design, full Markdown and LaTeX rendering, hands‑free voice/video calling (using local Whisper, OpenAI, Deepgram, Azure STT and Azure, ElevenLabs, OpenAI, Transformers, WebAPI TTS), built‑in key‑value artifact storage, and local RAG backed by nine vector databases (ChromaDB, PGVector, Qdrant, Milvus, Elasticsearch, OpenSearch, Pinecone, S3Vector, Oracle 23ai) with hybrid BM25+vector search, reranking, and multiple content‑extraction engines. Web‑search RAG pulls results from dozens of providers (SearXNG, Google PSE, Brave, Kagi, Mojeek, Tavily, Perplexity, Firecrawl, DuckDuckGo, Bing, etc.), while image generation and editing leverage DALL·E, Gemini, ComfyUI, and AUTOMATIC1111. Administrative dashboards track message volume, token consumption, and cost, and support model evaluation via arena matches, A/B testing, and ELO leaderboards. Storage options range from SQLite (with optional encryption) or PostgreSQL to S3, GCS, or Azure Blob, and enterprise authentication includes LDAP/AD, SSO via trusted headers/OAuth, and SCIM 2.0 provisioning. Observability is provided through OpenTelemetry, horizontal scalability via Redis‑backed sessions and WebSocket support, and multilingual i18n.
Read original
github-trending/python