A self-hosted ChatGPT alternative is a chat interface you run on your own server, connected either to a model provider's API with your own key or to a model running on your own hardware through Ollama. Open WebUI is the most popular choice and the easiest start. LibreChat is strong if you want many providers and agents in one place, and AnythingLLM is built around chatting with your documents.
None of these tools makes a model smarter than the one behind it. What they change is who holds your chat history, which models you can switch between, and how you pay.
How a self-hosted chat setup works
There are two parts. The interface is the web app: chat history, users, file uploads, prompts. The model is what answers. You can mix them freely.
With an API key, the interface sends your messages to a provider such as OpenAI, Anthropic, Google or Mistral, and you pay per use. Your chat history stays on your server, but the content of each request still goes to that provider.
With a local model, Ollama runs the model on your own machine, and nothing leaves it. Ollama has about 183,000 GitHub stars (October 2026), runs on macOS, Windows, Linux and in Docker, and is MIT-licensed. It also speaks a subset of the OpenAI API, so most chat interfaces can talk to it as if it were OpenAI. LocalAI is another option in the same role.
Open WebUI
Open WebUI is the most starred project in this space, with about 155,000 GitHub stars (October 2026), and it releases often. It supports "Ollama and OpenAI-compatible APIs", so you can put local models and cloud models side by side in one model picker. It is built to run fully offline, and it has multiple users, document chat (RAG), web search, tools and plugins.
Getting started is one Docker command, and the interface opens on port 3000. The documented Docker command is already set up to reach an Ollama running on the same machine.
A note on the license: code up to v0.6.5 is BSD-3-Clause. Since v0.6.6 (April 2025), the Open WebUI License adds a branding clause: you may not remove the "Open WebUI" name and logo, unless your deployment has 50 or fewer users in a 30-day period or you have an enterprise license. The project itself says the newer license is not OSI-approved. For a family or a small company this changes nothing. If you planned to rebrand it for a large customer base, read the license first.
Who it is for: almost everyone starting out, and especially people who want local models through Ollama.
LibreChat
LibreChat has about 46,000 GitHub stars (October 2026) and an MIT license. Its strength is breadth: Anthropic, AWS Bedrock, OpenAI, Azure OpenAI, Google and Vertex AI are built in, and Ollama and other OpenAI-compatible services work as custom endpoints.
On top of that come no-code agents, MCP support for tools, a sandboxed code interpreter, artifacts and web search. Logins work through OAuth2, LDAP or email, which makes it a good fit for a company where people should sign in with their existing accounts. It installs with Docker, npm or a Helm chart.
Who it is for: teams that want one interface for several paid providers, with proper user management.
LobeHub (formerly LobeChat)
LobeHub has about 83,000 GitHub stars (October 2026). It has moved from a chat app towards a platform for running AI agents, with an MCP marketplace and a polished interface. Self-hosting works with Docker Compose and a server database.
Check the license before you build on it. It is Apache 2.0 with extra conditions in the LobeHub Community License: commercial use of the unmodified software is allowed, but developing and distributing a derivative work requires a commercial license from LobeHub. Running it for yourself is fine.
Who it is for: people who like its interface and agent features and don't plan to ship a modified version.
AnythingLLM
AnythingLLM has about 67,000 GitHub stars (October 2026) and an MIT license. It focuses on chatting with your own documents: you put files into a workspace and ask questions about them. It comes as a desktop app for Windows, macOS and Linux, and as a Docker version.
The difference matters. The desktop app is single-user. Only the Docker version has user management, invitations, workspace permissions and embeddable chat widgets. For models it supports local runners such as Ollama, LM Studio and LocalAI, a built-in default, and cloud providers including OpenAI, Anthropic, Google Gemini, Mistral, Groq and OpenRouter.
Who it is for: anyone whose main use is asking questions about a pile of PDFs, manuals or notes.
Comparison table
| Open WebUI | LibreChat | LobeHub | AnythingLLM | |
|---|---|---|---|---|
| GitHub stars (Oct 2026) | about 155,000 | about 46,000 | about 83,000 | about 67,000 |
| Local models (Ollama) | Yes, first-class | Yes, custom endpoint | Check docs | Yes |
| Cloud providers | OpenAI-compatible APIs | Many built in | OpenAI and others | Many |
| Multi-user | Yes | Yes, OAuth2/LDAP/email | Check docs | Docker version only |
| Document chat | Yes | File search | Check docs | Main focus |
| License | Open WebUI License (BSD-3 based, branding clause) | MIT | Apache 2.0 + extra conditions | MIT |
What does it cost?
The software is free. What you pay for depends on the model side.
With API keys you pay the provider per token. For light personal use that is often less than a monthly chat subscription, and for heavy use it can be more. Every provider has a usage page, so check it after the first week instead of guessing. One advantage: you can use a cheaper model for simple questions and a stronger one only when you need it.
With local models you pay in hardware and electricity. The size of the model decides how much memory you need. Small models run on a normal laptop or a mini PC, but they are clearly weaker than the large cloud models. Large models need a graphics card with a lot of memory or an Apple Silicon machine with plenty of RAM. A cheap VPS without a GPU is fine for the chat interface, not for running a serious model.
A common middle way: run the interface on a small server, use cloud APIs for everyday work, and keep a local model for anything you'd rather not send out.
Privacy: what actually stays private?
With Open WebUI or LibreChat on your own server, your chat history, uploaded files and user accounts are stored by you. That is already a step up from a consumer chat account.
But if the model is a cloud API, each prompt and its attachments go to that provider. Read their data policy, especially whether API data is used for training and how long it is kept. Only a local model keeps the conversation entirely on your hardware.
How to run them
The usual route is your own server with Docker. Open WebUI is one container, LibreChat and LobeHub come with Compose files that bring their databases along. If you want local models, run Ollama on a machine with enough memory or a GPU, and point the interface at it. Self-hosting for beginners covers the basics of servers, domains and HTTPS.
If you only need the interface and plan to use API keys, a managed host is the easy option. On blitz.cloud, Open WebUI installs in one click from its App Store, hosted in the EU with HTTPS included. The free plan needs no credit card, and apps sleep after 30 minutes without visitors, which suits a chat you open a few times a day. A hosted interface like this will not run large local models for you, so pair it with an API key.
For more options, see ChatGPT alternatives. If you want AI to act on your other apps, Zapier alternatives and how to self-host n8n go further.
FAQ
Can I run ChatGPT itself on my own server?
No. The models behind ChatGPT are not available to download. You can use them through OpenAI's API in one of these interfaces, or run open-weight models locally with Ollama, including the gpt-oss models OpenAI has published.
What is the easiest self-hosted ChatGPT alternative?
Open WebUI. One Docker command gives you the interface, and it connects to an Ollama on the same machine without much setup.
Do I need a GPU?
Not for the interface, and not if you use cloud APIs. For local models a GPU or an Apple Silicon machine makes a big difference in speed and in which model sizes you can run.
Is a self-hosted chat cheaper than ChatGPT Plus?
It depends on how much you use it. With API keys you pay per use, which is cheap for occasional questions and can add up for long daily sessions. Watch your provider's usage page for a few weeks before you decide.
Is Open WebUI open source?
Versions up to v0.6.5 are BSD-3-Clause. Later versions use the Open WebUI License, which adds a branding requirement for deployments with more than 50 users in 30 days. The project says that license is not OSI-approved.