[email protected]

Open WebUI, LibreChat, AnythingLLM & Co. Compared 2026

Open WebUI, LibreChat, AnythingLLM & Co. Compared 2026

Editorial note: The information in this article was compiled to the best of our knowledge at the time of publication. Technical details, prices, versions, licensing terms, and external content may change. Please verify the information provided independently, particularly before making business-critical or security-related decisions. This article does not replace individual professional, legal, or tax advice.

Have an AI chat platform operated in Germany? WZ-IT takes over hosting and operations for Open WebUI, LibreChat, AnythingLLM, LobeChat, Onyx and PrivateGPT: updates, backups, monitoring and the connection to models. Open WebUI is also available as an AI platform with a Section 203 supplementary agreement. Book an initial consultation

Six open-source platforms are widely used for AI chat in the enterprise: Open WebUI, LibreChat, AnythingLLM, LobeHub, Onyx and PrivateGPT. They solve different tasks but look similar at first glance: an interface with language models and documents behind it. The difference lies in license, sign-in, permissions, knowledge base and operation.

This article compares the six platforms by these criteria and answers when switching away from Open WebUI is worthwhile. All figures come from the GitHub API, license files and vendor documentation as of 21 October 2026. Detailed comparisons of individual pairings are linked at the end.

Table of contents

  1. Short answer
  2. The six platforms in one sentence
  3. Basic data at a glance
  4. License and open-source status
  5. Open WebUI alternatives: when switching makes sense
  6. Operation and architecture
  7. Sign-in, SSO and permissions
  8. Knowledge base and connectors
  9. Models, agents and MCP
  10. Telemetry and data flow
  11. Security advisories of the last twelve months
  12. Vendor, support and cost
  13. Which platform for which use
  14. Out of scope: workflow builders
  15. Common misconceptions
  16. Data and methodology
  17. How WZ-IT works
  18. Further guides
  19. Sources

Short answer

Open WebUI LibreChat AnythingLLM LobeHub Onyx PrivateGPT
License Open WebUI License (BSD-3 plus branding clause) MIT MIT (repository, Docker); desktop app with its own terms of use LobeHub Community License (Apache 2.0 plus additional terms) MIT except ee directories Apache 2.0
Core purpose Chat platform for many users Multi-provider chat with agents Chat with documents, desktop and Docker Chat and agent platform Enterprise search with chat API layer for RAG
Sign-in OIDC, LDAP, SCIM, SAML via proxy OIDC, SAML, LDAP local accounts, Simple SSO Passthrough OIDC (Better Auth) OIDC, SAML; SCIM Enterprise only no user management
Knowledge base hybrid search, reranking RAG service with pgvector 9 vector databases pgvector, S3 hybrid index, 56 connector tiles Qdrant
Operation one container Compose with MongoDB one container Compose with PostgreSQL and S3 Compose or Helm, OpenSearch package or container, external inference server

Rule of thumb: Open WebUI as the central platform for many users, LibreChat for free branding and native connection of large providers, AnythingLLM for getting started via desktop app, Onyx for search across business systems with document permissions, LobeHub and PrivateGPT for teams that reuse the agent platform or the API respectively.

The six platforms in one sentence

  • Open WebUI: Chat interface with groups, access rights, hybrid RAG and broad sign-in options, originally created for Ollama.
  • LibreChat: Multi-provider chat with Agent Builder, MCP and native clients for Anthropic, Bedrock and Vertex AI; acquisition by ClickHouse announced in November 2025.
  • AnythingLLM: Chat with documents and agents as a desktop app for individuals and as a Docker server for small teams.
  • LobeHub (formerly LobeChat): Chat and agent platform with skill store, under its new name since version 2.0.
  • Onyx (formerly Danswer): Enterprise search with connectors to business systems, chat, agents and permission sync.
  • PrivateGPT: RAG service as an API layer following the Claude Messages schema; the interface is a demonstrator.

Basic data at a glance

Open WebUI LibreChat AnythingLLM LobeHub Onyx PrivateGPT
Repository open-webui/open-webui LibreChat-AI/LibreChat Mintplex-Labs/anything-llm lobehub/lobehub onyx-dot-app/onyx zylon-ai/private-gpt
Latest version 0.12.3 (16 Oct 2026) 0.8.8 (1 Oct 2026) 1.18.0 (13 Oct 2026) 2.2.20 (18 Oct 2026) 4.9.2 (15 Oct 2026) 1.0.1 (18 Jun 2026)
GitHub stars around 156,400 around 46,100 around 67,400 around 83,400 around 32,700 around 57,600
Vendor Open WebUI Inc. ClickHouse (acquisition announced 4 Nov 2025) Mintplex Labs Inc. LobeHub LLC DanswerAI, Inc. Zylon
Archived no no no no no no

On versions: GitHub marks all LibreChat releases as pre-release; the latest non-RC tag is treated as current. Alongside 4.9.2, Onyx publishes the pre-release 5.0.0-beta.2 (20 October 2026). PrivateGPT has only 11 commits on the main branch between September 2024 and May 2026; version 1.0.0 of 3 June 2026 is a rewrite.

License and open-source status

Platform License Additional terms
Open WebUI Open WebUI License Branding must not be removed or replaced, except with at most 50 end users in 30 days, written permission or an enterprise license; since version 0.6.6
LibreChat MIT none; the Admin Panel is separately under AGPL-3.0
AnythingLLM MIT MIT applies to the repository, source code and Docker version; the desktop app has its own terms of use (as of 14 Jul 2025: personal and internal business purposes only, no redistribution, no derivatives, no SaaS); Pro subscription and cloud are paid
LobeHub LobeHub Community License Commercial use as a frontend and backend service without source code changes is permitted; a commercial license from the vendor is required to develop and distribute derivative works. Contributors agree that the vendor may make the license stricter or looser and use contributions commercially.
Onyx MIT, ee directories under Onyx Enterprise License Production use of the ee parts only with a subscription
PrivateGPT Apache 2.0 none; enterprise features in the separate product Zylon

GitHub shows no standard license for Open WebUI, LobeHub and Onyx (NOASSERTION, "Other"). For LobeHub, the README badge shows Apache 2.0; the license text contains the additional terms. The history: LobeHub changed from MIT (2023) to Apache 2.0 with additional terms (17 June 2024) and renamed the license on 21 September 2025; Open WebUI changed from MIT via BSD-3-Clause to the Open WebUI License. The series of license changes in the industry is covered in the chronicle of open-source license changes, the branding clause of Open WebUI in the article Open WebUI License and Branding.

Open WebUI alternatives: when switching makes sense

Situation Alternative Reason
Own branding or operation under your own brand without a vendor license LibreChat or AnythingLLM (Docker version) MIT without a branding requirement
Native connection of Anthropic, Bedrock or Vertex AI LibreChat native clients; Open WebUI uses the OpenAI-compatible interfaces
SAML without an upstream proxy LibreChat or Onyx SAML in the application
Search across Confluence, SharePoint, Jira and Slack with document permissions Onyx permission sync (license part of the Enterprise edition) for 14 connectors according to the docs (4.9.2; 5.0.0-beta.2: 17)
Getting started on a laptop without a server AnythingLLM Desktop desktop app for Windows, macOS, Linux; observe the terms of use of the desktop app
Building your own application on a RAG API PrivateGPT API layer without its own interface for end users
Agent and skill store LobeHub skill store with community skills; observe the license terms for derivative works
Groups, SCIM and access per model and knowledge base stay with Open WebUI documented in this breadth only there without an enterprise license

For the detailed comparisons: Open WebUI vs LibreChat, Open WebUI vs AnythingLLM and Open WebUI vs Onyx.

Operation and architecture

Required and optional services Minimum values per docs Kubernetes
Open WebUI none (SQLite, embedded ChromaDB); multiple instances need PostgreSQL, vector database, Redis not stated Helm chart
LibreChat MongoDB; in the standard Compose additionally Meilisearch (chat search) and RAG service with PostgreSQL (file search); Redis for multiple instances 1 GiB RAM, 1 vCPU; Docker Compose stack with all services at least 2 GB official Helm chart
AnythingLLM none (SQLite, LanceDB) 2 GB RAM, 2 cores with AVX2, 5 GB Helm chart in the repository, appVersion 1.85.0
LobeHub PostgreSQL 14+; pgvector for the knowledge base; S3-compatible storage for uploads and the knowledge base (recommended by the docs); one model provider 2 cores, 4 GB RAM, 20 GB not found in the repository
Onyx Standard: PostgreSQL, OpenSearch, Redis, model server, SeaweedFS (replaceable via Compose profile and FILE_STORE_BACKEND=postgres); Lite: 5 services Lite 2 vCPU, 2 GB; Standard 4 vCPU, 10 GB RAM Helm chart, Terraform modules
PrivateGPT Qdrant (local mode, external in production), external OpenAI-compatible inference server not stated not in the repository

None of the platforms runs language models itself by default; exceptions are the bundled Ollama image of Open WebUI and the built-in engine of AnythingLLM (desktop; optional in Docker from the release after 1.18.0). Embedding, reranking and Whisper models run locally in Open WebUI (CPU by default, GPU with the :cuda image); Onyx runs model servers for embeddings. GPU sizing for language models depends on the model server, not on the interface.

Sign-in, SSO and permissions

Open WebUI LibreChat AnythingLLM LobeHub Onyx PrivateGPT
Local accounts yes yes, with passkeys yes (Docker) yes, magic link yes no (optional static secret)
OIDC yes yes not described; Simple SSO Passthrough (token-based one-time login link via API key) yes, many providers yes no
SAML via proxy yes not described not described yes no
LDAP yes yes not described not described not described only in Zylon
SCIM yes not found not described not described Enterprise only no
Groups yes, nested via Admin Panel, Entra ID sync no not described from 4.7, custom groups Enterprise only no
Access per knowledge source ACLs agents via ACL workspace assignment not described permission sync (Enterprise edition, license part ee) no
Audit log configurable, off by default yes, admin audit log (permissions, roles, groups, shares, configuration) with hash chain and CSV export since 0.8.7; in the docs only mentioned in the changelog event logs not described JSON event stream, always on according to the docs only in Zylon

For Onyx, the vendor statements contradicted each other (documentation: one SSO provider on every plan; enterprise page and pricing page: OIDC and SAML as an Enterprise feature); according to the source code, one provider is usable without a license, while several simultaneous providers require Business and SCIM requires Enterprise. "Not described" means that the documentation does not mention the feature.

Knowledge base and connectors

Search Vector database Connecting sources
Open WebUI Hybrid of BM25 and vector, reranking 13 according to docs and code (README states 9); default ChromaDB Google Drive, OneDrive/SharePoint; oikb with 40+ sources
LibreChat Vector search, full-text mode; reranking only in web search pgvector, Atlas MongoDB SharePoint Online, OneDrive for Business
AnythingLLM Vector search; reranking only with LanceDB 9 (default LanceDB) Confluence, Drupal Wiki, Obsidian, Paperless-ngx, GitHub/GitLab/Gitea, YouTube, websites
LobeHub Vector search; hybrid and reranking not described pgvector agent connectors to Google Drive, OneDrive, Dropbox and Notion (access via tool); Notion import via ZIP; no synchronization into the knowledge base described
Onyx Hybrid index of vector and keyword, optional reranking OpenSearch 56 connector tiles (Confluence, SharePoint, Notion, Jira, Slack etc.)
PrivateGPT Semantic search; keyword and hybrid search are provided for in the API schema but not yet implemented according to the API reference (HTTP 501) Qdrant only in Zylon

The connector count of Onyx varies by source (40+ on the pricing page, 50+ in the README, 56 tiles in the docs overview). For sources with file permissions, the permission check belongs before the vector search; how this works with Nextcloud is shown in Connecting Open WebUI to Nextcloud.

Models, agents and MCP

Models Agents MCP
Open WebUI Ollama, OpenAI-compatible; Anthropic, Gemini, Bedrock via compatible interfaces Agent models, Filters, Actions, Pipes, Tools, Skills Client, administrators only
LibreChat native: Anthropic, Bedrock, OpenAI, Azure, Google, Vertex AI; custom endpoints Agent Builder, Marketplace, Skills Client (stdio, SSE, Streamable HTTP)
AnythingLLM 38 providers according to the README @agent skills, Agent Flows Client (not in the vendor cloud)
LobeHub 57 provider guides Agent Builder, Agent Groups, Skill Store Client, community MCP integrations
Onyx OpenAI, Anthropic, Azure, Bedrock, Vertex AI, Ollama, LiteLLM, vLLM etc. Custom Agents, Deep Research, Code Interpreter Client and server
PrivateGPT any OpenAI-compatible server Tool use, Skills (basic) Client, also remote MCP

The code execution of LobeHub runs by default in a sandbox of the vendor (market.lobehub.com); a self-operated service (Onlyboxes) is the alternative. In Open WebUI, Pyodide is now considered legacy; Open Terminal is recommended.

Telemetry and data flow

Default Switching off
Open WebUI Docker image: telemetry switched off via SCARF_NO_ANALYTICS=true, DO_NOT_TRACK=true, ANONYMIZED_TELEMETRY=false; update check on OFFLINE_MODE
LibreChat no vendor telemetry in documentation, example configuration and Compose files -
AnythingLLM Telemetry on (PostHog); connections to cdn.anythingllm.com and GitHub DISABLE_TELEMETRY=true
LobeHub Analytics only via environment variable (Vercel deployments); connections to market.lobehub.com empty the variables
Onyx Telemetry on (telemetry.onyx.app); from 5.0 beta, Fleet Telemetry replaces the previous anonymous telemetry (heartbeat every 5 minutes, including connector counts and email domains of sign-ins) DISABLE_TELEMETRY=true
PrivateGPT no telemetry code found -

Security advisories of the last twelve months

Published GitHub security advisories from 21 October 2025 to 21 October 2026:

Total critical high Focus areas
Open WebUI 163 1 70 XSS, sign-in, SSRF, hangs
LibreChat 28 3 9 MCP, SSRF, XSS
AnythingLLM 23 1 5 multi-user permissions, RCE
LobeHub 7 1 0 XSS to RCE, SSRF, sign-in bypass
Onyx 4 1 0 token leak via MCP, IDOR
PrivateGPT 0 0 0 -

The number says nothing about the quality of a project. It shows how much was reported and published, and depends on size, attack surface and reporting culture; the public PrivateGPT repository shows hardly any changes between September 2024 and May 2026 (11 commits on the main branch). For all of them: apply updates promptly, release MCP servers and tools to administrators only, and do not expose the instance to the internet without sign-in.

Vendor, support and cost

Self-hosting Commercial offering by the vendor
Open WebUI free with unchanged branding Enterprise license (white-label, Terminals), price not public
LibreChat free (MIT) no public enterprise or support offering; ClickHouse features LibreChat as the chat interface of its open-source "Agentic Data Stack"; a separate paid LibreChat product is not documented
AnythingLLM free (MIT; desktop app with its own terms of use) Cloud: Basic USD 50, Pro USD 99 per month; Enterprise on request
LobeHub free under the Community License Cloud: Free, Starter USD 9.90, Premium USD 19.90, Ultimate USD 39.90 per month with annual billing; Enterprise on request
Onyx free (MIT part) Business USD 20 per user and month with annual billing, including permission sync according to the pricing page; Enterprise on request
PrivateGPT free (Apache 2.0) Zylon, price not public

Vendor prices: as of 21 October 2026, in US dollars according to the pricing pages. The operating costs of self-hosted platforms arise from hosting, updates, backups, monitoring and the connection of the models.

Which platform for which use

Use Best fit To note
Central platform for many users with SSO, groups, access per model Open WebUI Branding clause; multi-instance operation needs PostgreSQL, Redis, vector database
Multiple providers natively, MCP agents, MIT LibreChat MongoDB; no SCIM; beta features in 0.8.8
Small group, getting started via desktop app AnythingLLM no groups, no OIDC/LDAP/SAML in the docs, local accounts and Simple SSO Passthrough; desktop app with its own terms of use
Search across business systems with document permissions Onyx Permission sync as a license part of the Enterprise edition; standard mode with OpenSearch
Agent platform with skill store LobeHub License for derivative works; sandbox at the vendor
RAG API for custom applications PrivateGPT no user management; 1.0 is a rewrite

Out of scope: workflow builders

Dify, Flowise, Langflow and n8n do not belong in this comparison. They build workflows and agents instead of providing employees with a chat interface. License and hosting for third parties differ greatly there: Flowise has been archived since August 2026, n8n is under the Sustainable Use License. The comparison is provided by Dify, Flowise, Langflow, n8n compared.

Common misconceptions

  • "Open WebUI, LibreChat and AnythingLLM are all MIT." No. Only LibreChat and AnythingLLM (for AnythingLLM, the repository, source code and Docker version; the desktop app has its own terms of use). Open WebUI is under the Open WebUI License with a branding clause.
  • "LobeChat no longer exists." The project has been called LobeHub since version 2.0; the old repository redirects.
  • "PrivateGPT is the simple chat app with your own documents." That corresponds to version 0.6.2. Version 1.0 is an API layer.
  • "Onyx is entirely MIT." Only outside the ee directories. Permission sync, SCIM and custom groups are in the Enterprise edition (license part ee).
  • "Self-hosted means no connection to the outside." Open WebUI checks for updates; AnythingLLM and Onyx send telemetry until this is switched off.
  • "More GitHub stars means better for enterprises." The stars differ by a factor of five (Open WebUI 156,400, Onyx 32,700) and say nothing about sign-in, permissions or license.

Data and methodology

As of: 21 October 2026. The data basis is the GitHub API (license, version, stars, published security advisories), the LICENSE files and individual source files of the repositories, and the vendors' documentation. Stars are rounded and change daily. "Not described" and "not found" mean that documentation and repository structure do not mention the feature; the function may exist in the code. The LobeHub documentation website was not complete when retrieved; the information comes from the documentation source files in the repository. The advisories cover the period from 21 October 2025 to 21 October 2026. Contradictions in the vendor statements (Onyx: SSO plan, resolved by the source code, and connector count; Open WebUI: number of vector databases) are noted.

Tables and figures may be reused with attribution and a link to this article.

How WZ-IT works

WZ-IT runs Open WebUI, LibreChat, AnythingLLM, LobeChat, Onyx and PrivateGPT as a hosting and operations offering in Germany: application, databases, proxy, updates, monitoring and backup, with technically secured credentials and model endpoints. The customer decides which models, user groups and data sources are permitted. Open WebUI runs with visible original branding and is available as an AI platform with models operated in Germany by German providers, SSO and a Section 203 supplementary agreement from EUR 149.90 net per month; we clarify license questions on number of users and branding before rollout. For Onyx, WZ-IT operates the freely licensed edition (MIT outside the ee directories); features under the ee directories are not part of the offering, and we support the procurement of a vendor license on request. For operation in your own network, the platform is available on the App Node and, with its own GPU, on the AI Cube. Models from Germany are provided by LLM hosting.

Further guides

Choose an AI chat platform or have it operated? We assess the number of users, sign-in, sources and models and run the platforms in Germany. Book an appointment

Sources

Timo Wevelsiep

Written by

Timo Wevelsiep

Co-Founder & CEO

Co-Founder of WZ-IT. Specialized in cloud infrastructure, open-source platforms and managed services for SMEs and enterprise clients worldwide.

Frequently Asked Questions

Answers to important questions about this topic

Companies worldwide trust WZ-IT

  • Stadtwerke Brühl
  • DGHO e.V.
  • ABCO Water Systems
  • Golem.de
  • EVADXB
  • nextGYM
  • AInergy
  • ml&s
  • Odiseo Solutions
  • Annota
  • ARGE
  • SweetConnect GmbH
  • Aphy AG
  • CORGOS
  • Rekorder
  • SolidProof
  • Yonju
  • Keymate
  • Paritel
  • Mr. Clipart
  • Millenium
  • Negosh
  • Führerscheinmacher
  • Boese VA
Read client reviews

Enquiry

Choose an AI chat platform or have it operated

We run Open WebUI, LibreChat, AnythingLLM, LobeChat, Onyx and PrivateGPT in Germany and assess the number of users, sign-in, sources and models.

  • Straight with Timo and Robin - no sales team, no pitch
  • A professional assessment of your situation
  • Concrete next steps for infrastructure, software or AI

The conversation is free and non-binding.

WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.
Jakob ÖschlbergerInno7 GmbH
Timo and Robin, founders of WZ-IT

What are you planning?

WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.
Jakob ÖschlbergerInno7 GmbH