LobeHub vs Open WebUI

LobeHub and Open WebUI are two self-hosted chat interfaces that put a ChatGPT-style window in front of whatever models you can reach, hosted or running on your own hardware. Both have moved past plain chat into agents, tools and document search, and both publish an official Docker setup. The differences are in the deployment shape, how each one talks to model providers, who can sign up by default, and what each licence lets you change. LobeHub, formerly LobeChat, is a Next.js app with PostgreSQL, Redis and S3 storage beside it. Open WebUI is a Python app in one container.

Two shapes of deployment

Open WebUI runs as a single container. The quick start is one docker run with a named volume at /app/backend/data. The README calls that volume "crucial" because it holds the database. Storage defaults to SQLite, with PostgreSQL as the option for larger installs, and files can go to local disk, S3, Google Cloud Storage or Azure Blob Storage. An :ollama image bundles Ollama in the same container, and a :cuda image adds GPU support. The quick start we list pulls the :main tag rather than a pinned version.

LobeHub's self-hosted server edition is a Compose stack. Its setup.sh script downloads the Compose file and writes a .env with generated secrets. The stack is the LobeHub app, an agent gateway, PostgreSQL (the ParadeDB image, which carries pgvector and pg_search), Redis, RustFS for S3-compatible file storage, and SearXNG for web search. Elasticsearch is available as an opt-in Compose profile. The app listens on 3210. setup.sh points the S3 endpoint (RustFS, port 9000) and the agent gateway (port 8787) at your public host, so browsers must be able to reach those ports too.

Two things to fix before exposing a LobeHub box. First, the upstream Compose file publishes PostgreSQL on 5432 and Redis on 6379 to the host, so firewall them. Second, the setup script generates the auth, key-vault, S3 and gateway secrets, but we found no step that changes POSTGRES_PASSWORD from the value in the .env template. Set your own before the first start.

Models and providers

Open WebUI is built around two connection types: local Ollama and any OpenAI-compatible API. The README suggests pointing that API URL at LM Studio, GroqCloud, Mistral, OpenRouter, vLLM "and more". Ollama is the smoothest path. It can run on the same host, on another server via OLLAMA_BASE_URL, or inside the bundled image.

LobeHub documents a native environment variable for about forty providers, including OpenAI, Anthropic, Google, Azure, DeepSeek, xAI, Mistral, Groq and OpenRouter. Ollama is enabled with OLLAMA_PROXY_URL. If most of your traffic goes to hosted providers, LobeHub has less to configure. If it goes to a model on your own hardware, Open WebUI's defaults already point there.

Features beyond chat

Open WebUI has built-in RAG: hybrid search (BM25 plus vectors) with reranking, nine vector database options (ChromaDB, PGVector, Qdrant, Milvus and others), several content-extraction engines, and web search through a long list of providers including SearXNG and Brave. It is extended through Filters, Actions, Pipes, Tools and Skills, and it connects to MCP and OpenAPI tool servers. The README also lists Notes, Channels, persistent memory, calendars and scheduled automations.

LobeHub now puts agents first. Its README lists an Agent Builder, Agent Groups that work on a task together, scheduled runs, projects, shared workspaces, and structured, editable "White-Box Memory". It claims "10,000+" tools and MCP-compatible plugins. Its knowledge base stores vectors in PostgreSQL with pgvector and uploaded files in S3. Embeddings default to OpenAI's text-embedding-3-small, and the docs name Ollama among the alternative embedding providers. On a server with no OpenAI key, change the embedding model before you rely on document search.

Users and sign-in

Open WebUI promotes the first account to administrator, "and sign-up then closes". Admins manage roles, groups and per-group permissions. For company directories it supports LDAP and Active Directory, SSO through OAuth providers or trusted headers, and SCIM 2.0 provisioning.

LobeHub signs users in through Better Auth: email and password, and SSO providers including Google, GitHub, Microsoft, Keycloak, Authentik, ZITADEL, Okta and any generic OIDC server. Registration is open by default. The .env template says to leave AUTH_ALLOWED_EMAILS empty "to allow all users to register". On a public address, set that allow-list, or AUTH_DISABLE_EMAIL_PASSWORD=1 for SSO-only sign-in.

Licences

Neither is OSI-approved open source. Both are source-available.

Open WebUI uses the Open WebUI License, which is BSD-3-Clause with a branding clause. You may not alter or remove the "Open WebUI" branding unless your deployment has no more than 50 end users in any rolling 30-day period, you have written permission, or you hold an enterprise licence.

LobeHub uses the LobeHub Community License, which is Apache 2.0 with added conditions. It may be used commercially "as a frontend and backend service without modifying the source code". To "develop and distribute a derivative work", you need a commercial licence.

For an internal deployment that you do not modify, neither licence gets in the way. They restrict different things: Open WebUI limits rebranding, and LobeHub limits distributing modified versions.

Footprint

Both installs have been tested on this site on Ubuntu 26.04. Idle after first start on our GCP e2-standard-2 test box, LobeHub used ~718 MB of RAM (2026-09-29) and Open WebUI used ~928 MB (2026-09-28). We found no published RAM minimum for either, so our 2 GB floor for each is an estimate. Any local model you run beside them needs its own memory on top. We rate both 2 out of 5 to deploy. Open WebUI is one container. LobeHub is a script plus a Compose stack, but the script does most of the work.

Which to choose

Choose Open WebUI if your models run on Ollama or another OpenAI-compatible server, you want one container with built-in document search, and you need directory sign-in (LDAP, SCIM) or fine-grained roles for a team. See the Open WebUI deploy guide and Open WebUI with Ollama.

Choose LobeHub if you mostly use hosted providers and want each one configured natively, and if agents, agent groups and scheduled runs matter more than a single container. Plan for PostgreSQL, Redis and S3 storage, an embedding model for the knowledge base, a new database password, and a sign-up allow-list.

Choosing between Open WebUI and LobeHub

Starting from Open WebUI, the usual reason to move is providers or agents: native Anthropic or Google connections without a proxy, or agent groups that work on a task together. Starting from LobeHub, the usual reason to move is local models and simpler operations, with one container and Ollama as the default backend. For other options, see Open WebUI vs LibreChat and AnythingLLM vs Open WebUI.