Self-hosted AI assistant for a team

Open WebUI leads with a score of 89. Updated from public GitHub data.

Full category: Chat UIs.

Which one?

1st pickOpen WebUI
Self-hosted web UI for Ollama and OpenAI-compatible models with users and RAG
Score 89 · 159k stars · custom license
Switch togradio
Python library for building web UIs around ML models and functions
for a permissive license (Apache-2.0) · Score 89 · 44.1k stars · Apache-2.0
Switch toOnyx
Team knowledge search and chat that indexes 50+ connected apps with permissions
to try before you install (live demo) · Score 88 · 33.4k stars · custom license
Self-hosted AI assistant for a team
RankProjectAdoptionFreshnessMaintenanceAgent-readyScore
1Open WebUISelf-hosted web UI for Ollama and OpenAI-compatible models with users and RAG159k starscustom licenselast commit Nov 202610010097089 out of 100
2gradioPython library for building web UIs around ML models and functions44.1k starsApache-2.0last commit Nov 202692100904589 out of 100
3OnyxTeam knowledge search and chat that indexes 50+ connected apps with permissions Live demo ↗ (opens in a new tab)33.4k starscustom licenselast commit Nov 202680100858588 out of 100

Momentum, Verified build, Docs and Privacy are not measured yet; their weight goes to the signals shown. A dash means the signal is not scored for that kind of project. Hover a number for its rating in words.

Reviews

189 out of 100

Open WebUI

Self-hosted web UI for Ollama and OpenAI-compatible models with users and RAG

159k starscustom licenselast commit Nov 2026

Open WebUI is a self-hosted chat front-end for Ollama and any OpenAI-compatible API, such as vLLM, LMStudio, OpenRouter or Groq. It adds role-based access control, chat history, document RAG with 9 vector database options, web search, voice calls, image generation and a plugin system (Filters, Pipes, Tools, MCP). It installs via pip, Docker or Kubernetes and serves on port 8080 by default.

Strengths

  • Works with Ollama and any OpenAI-compatible endpoint, so local and cloud models mix in one UI
  • RBAC, user groups, LDAP, OAuth/SSO and SCIM 2.0 provisioning for multi-user deployments
  • RAG with hybrid BM25 plus vector search and 9 vector databases, including ChromaDB, PGVector and Qdrant
  • SQLite or PostgreSQL, S3/GCS/Azure storage, Redis and OpenTelemetry for scaled deployments

Weaknesses

  • Custom license requires preserving Open WebUI branding; not a standard OSS license
  • Per-user isolated terminals are in the Enterprise edition only
  • Very broad feature set, so the many settings take time to configure
  • Does not run models itself; needs Ollama or an API unless using the :ollama image
  • GPU optional
  • Docker + Compose
  • Needs Ollama or OpenAI-compatible API, Docker (optional), PostgreSQL (optional), Redis (optional)
  • Models: Ollama, OpenAI-compatible APIs, vLLM, LMStudio, GroqCloud
  • port 8080
289 out of 100

gradio

Python library for building web UIs around ML models and functions

44.1k starsApache-2.0last commit Nov 2026

Gradio wraps any Python function in a browser UI using Interface, Blocks for custom layouts, or ChatInterface for chatbots, with more than 30 built-in components. It can create a temporary public gradio.live URL via share=True while computation stays on the local machine. Python and JavaScript clients can query running apps programmatically.

Strengths

  • Few lines of Python give a working UI; no JavaScript or CSS needed
  • share=True creates a public gradio.live link while compute stays local
  • Python and JavaScript clients can call any Gradio app programmatically
  • Hot reload mode via `gradio app.py` speeds up iteration

Weaknesses

  • Blocks layouts are limited to Gradio's component set unless you write custom components
  • Not a hosted chat product; no user accounts or chat history described
  • Requires Python 3.10 or higher
  • Shared links depend on Gradio's relay service rather than your own hosting
  • no GPU
  • Needs Python 3.10+
  • Models: any Python function, Hugging Face models
  • port 7860
388 out of 100

Onyx

Team knowledge search and chat that indexes 50+ connected apps with permissions

Live demo ↗ (opens in a new tab)33.4k starscustom licenselast commit Nov 2026

Onyx indexes content from 50+ applications along with metadata and permissions, then serves it through a chat UI with agentic RAG, deep research, custom agents, web search, MCP actions and a code sandbox. It works with self-hosted and hosted LLM providers such as Ollama, vLLM, LiteLLM, OpenAI, Anthropic and Gemini. A Lite mode runs the chat and agents without document indexing.

Strengths

  • Indexes 50+ connectors with source permissions, so answers respect existing access controls
  • Hybrid vector and keyword index plus a retrieval-tuned agent for search
  • Works with Ollama, vLLM, LiteLLM and major hosted LLM providers
  • Exposes an MCP server, Slack and Discord bots, Chrome extension and embeddable widget

Weaknesses

  • Standard mode needs a multi-service stack: index, workers, Redis, MinIO and model inference servers
  • SSO, SCIM, RBAC and analytics are in the Enterprise Edition
  • Lite mode cannot index documents
  • License detected as NOASSERTION; README claims MIT for Community Edition only
  • GPU optional
  • Docker + Compose
  • Needs Redis, MinIO, vector and keyword index, Docker or Kubernetes
  • Models: Ollama, vLLM, LiteLLM, OpenAI, Anthropic