.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
Updated and tested 30 September 2026
LibreChat has been a ClickHouse product since the acquisition was announced on 4 November 2025. The licence is still MIT and releases have kept coming; the announcement also promises continued investment in the open-source project, so the ownership change on its own isn't a reason to leave. The reasons that do come up are more ordinary. Mostly the six-service compose stack is more than a small team uses, and MongoDB is a database nobody else on the server wanted. Sometimes the need was only a chat window for Ollama, and an agent platform is overkill for that. Eight LibreChat alternatives follow, ranked by how much of its coverage each one keeps. None of them need MongoDB.
Six containers for one LibreChat install on my server, with the admin panel moved to 3001 because Open WebUI had 3000.
Alternatives with logins and shared history
Open WebUI keeps the most (the Open WebUI vs LibreChat comparison goes through it in detail). One container, SQLite by default, built-in RAG with a choice of 13 vector backends, groups with per-model permissions, OAuth, LDAP and SCIM provisioning, native MCP over Streamable HTTP and a model builder that stands in for LibreChat's agents, all of it set up in the Open WebUI with Ollama install guide. What it drops: native endpoint types for Anthropic, Google, Bedrock and Vertex (everything non-Ollama goes through the OpenAI-compatible format, with a proxy or Pipe function for the rest), SAML and per-user token balances. In place of LibreChat's code-interpreter service it runs Python in the browser (Pyodide) or in Open Terminal, which sits in its own container. The licence is the other difference: since v0.6.6 it carries a branding clause for installs above 50 users in a 30-day period, which matters only if you rebrand it, and an enterprise licence from the Open WebUI team removes it. On my test server it also closed sign-ups by itself once I had created the first admin account.
Open WebUI's Authentication settings after my first account: New Sign Ups off and pending as the default role.
AnythingLLM (MIT) keeps the agent side better than most and is stronger than LibreChat on documents. Workspaces bundle files, embeddings and chat together; the Docker build has multi-user accounts and permissions; there's a no-code agent builder with web browsing and MCP support; and LanceDB ships inside the container so RAG works with no second database. It loses LibreChat's breadth of native providers (though close to 40 are listed, including Ollama and LM Studio) and has no SAML.
LobeHub is the closest in ambition. Its LobeHub Community License is Apache 2.0 with added conditions: commercial use is fine unmodified, and a distributed derivative needs a commercial licence. Scheduled agents, groups of agents working together, a marketplace of plugins, MCP and a built-in knowledge base. The self-hosted database mode wants PostgreSQL with pgvector and S3-compatible storage (logins are built in, SSO optional), so it trades MongoDB for a stack of similar size. The reasons to pick it are the interface and the plugin catalogue. A lighter install isn't one of them.
Onyx (MIT for the Community Edition) keeps agents, MCP, web search and code execution and adds something LibreChat doesn't have: 50-plus connectors that index Slack, Confluence, Google Drive and similar into a searchable knowledge layer. The standard deployment is heavier than LibreChat's, with a vector database, Redis, MinIO and workers, though a lite mode exists. Consider it only if the chat has to answer from what's already in the company wiki.
Single-user and desktop options
The options below have no multi-user features, and multi-user is most of why people run LibreChat in the first place. They suit one person who wants the chat window and a plain licence.
big-AGI (MIT) is the interesting one. Its Beam feature sends the same prompt to several models side by side, it has personas, voice, image generation and 20-plus providers, and it deploys with Docker or on a serverless platform. Storage is the browser, so there's no server-side history and no accounts; the project page sends teams that need shared prompts or SSO to the maintainers. NextChat (MIT) is the same shape with a shared-password gate and an optional sync backend, and it's been around long enough that the Docker image is everywhere.
Jan (Apache 2.0) and Cherry Studio (AGPL-3.0) are desktop apps rather than servers, and the Open WebUI alternatives roundup covers the wider set of desktop and browser-only tools, Hollama and Ollama's own app included. Jan runs GGUF models itself through llama.cpp and exposes an OpenAI-compatible server on port 1337; Cherry Studio connects to providers and Ollama and adds custom assistants and knowledge bases. Both support MCP. Neither has any concept of a second user.
