Best open source chat UIs for LLMs

Why run your own chat UI?

A self-hosted chat UI gives your team a ChatGPT-style interface while you decide which models it calls and where conversations are stored. That matters for privacy, compliance and cost control.

It also lets you mix models: a local model for sensitive data, a frontier API for hard problems, all in one place with shared prompts and document collections.

A shared UI also turns individual prompt tricks into team assets. Presets, shared prompts and document collections mean a good workflow discovered by one person reaches everyone.

Finally, it gives you one place to see usage. You learn which tasks people actually use AI for, which is the best input for deciding what to automate next.

Which open source chat UIs are worth trying?

When is Open WebUI the right pick?

Open WebUI started as a front end for Ollama and grew into a full platform with user management, document upload, web search, tool calling and custom pipelines. If your models run on Ollama or any OpenAI-compatible server, it connects with little effort.

It is a strong default for teams that want local models first. Read the current licence file before you rebrand or resell it, because the project has added terms around branding.

Its pipelines feature lets you insert custom Python between the user and the model, for filtering, logging or routing. That is useful once basic chat is working and you want organisation-specific behaviour.

When should you pick LibreChat or LobeChat?

LibreChat is built for multi-provider use with proper accounts. It supports many hosted APIs side by side, conversation search, presets and agent features, and it integrates with common single sign-on providers. It suits an internal company assistant.

LobeChat puts more weight on interface polish and extensibility through plugins and agent definitions. It can run as a simple client-side app or with a server database for multi-device sync.

Both move quickly, so compare the current feature lists yourself rather than relying on older reviews.

When is AnythingLLM or Jan the better fit?

AnythingLLM organises chats into workspaces, each with its own documents and settings. Upload files, it embeds them, and the chat answers from that knowledge with citations.

It is the quickest route to “chat with our documents” for a non-technical team. If you need deep control over chunking and retrieval, a dedicated RAG framework behind a simpler UI may serve you better.

Jan targets a different user: one person on one laptop who wants models to run fully offline. It is a good recommendation for consultants handling sensitive files, but not a shared server.

How to choose a chat UI

• Count users. One person can use Jan or a desktop app; a team needs accounts, roles and SSO.
• List model sources: local Ollama or vLLM, cloud APIs, or both.
• Decide how important document chat is versus plain conversation.
• Check where conversations and files are stored, and how you back them up.
• Review the licence if you plan to brand, embed or resell the UI.
• Try the top two for a week with real users before standardising.

Common mistakes when self-hosting a chat UI

• Exposing the UI to the internet without TLS, strong auth or disabling open sign-up.
• Letting every user paste API keys instead of managing keys centrally.
• No backups of the database holding conversations and uploaded files.
• Assuming built-in RAG handles complex documents well without testing it on your files.
• Pinning to a months-old version of a fast-moving project and missing security fixes.

What sits behind the chat UI

The UI is only the front door. Behind it you need a model server, possibly a gateway for keys and budgets, and a vector store for documents. Keep those layers separate so you can swap the UI later without losing data or model setup.

RepoLoot’s catalog covers the surrounding pieces as well, such as model servers and gateways, tagged by difficulty so you can see what a full stack takes before you start.

Export conversations and documents periodically in an open format. Chat UIs change quickly, and a clean export keeps you free to move if a project stalls or changes its licence.

How do these chat UIs handle documents and RAG?

Most of these projects include some form of document chat, but the depth varies. Built-in RAG usually means upload, automatic chunking, embedding with a default model and a simple similarity search.

That works for policies, manuals and notes. It struggles with scanned PDFs, complex tables and large collections that need filtering by department or date. Test with your worst documents, not your cleanest.

How do you roll a chat UI out to a team?

Start with a small pilot group and one or two approved models. Too many models at once confuses users and makes costs hard to predict.

Connect sign-in to your identity provider, disable open registration, and set default models per group. Share a handful of tested prompts or presets for common tasks, such as summarising meetings or drafting replies, so people see value on day one.

Write a short usage policy: what data may be pasted, which models are approved for client data, and who to ask for help.

How do these projects compare on operations?

Features get the attention, but running cost comes from operations: how many services you deploy, what needs backing up and how painful upgrades are.

Open WebUI can start as a single container with an embedded database and grow to external databases later. LibreChat typically runs alongside MongoDB, a search service and an optional RAG API, so plan a Compose stack from day one. LobeChat is lightest in client-only mode but needs a database and auth provider for real multi-user use.

AnythingLLM’s Docker image bundles most of what it needs, which keeps a small deployment simple. Jan runs on the desktop and needs no server at all.

• Back up the database and the uploads directory together
• Read release notes before upgrading; schema migrations are common
• Keep one staging instance to test upgrades on a copy of real data

Frequently asked questions

What is the best open source ChatGPT-style interface?

Open WebUI is the most widely used, particularly with local models through Ollama. LibreChat is a better fit when you mix many cloud providers with user accounts, and AnythingLLM is best when chatting with documents is the main goal. Try two with real users before choosing.

Can these chat UIs work with Claude and GPT models?

Yes. LibreChat, LobeChat, Open WebUI and AnythingLLM can connect to major hosted APIs, either natively or through an OpenAI-compatible endpoint or gateway. You supply your own API keys, and usage is billed by each provider.

Is Open WebUI free for commercial use?

You can generally self-host it inside a company, but the project’s licence includes conditions, including around branding, that have changed over time. Read the current licence file in the repository before rebranding it or offering it to customers.

Do I need a GPU to run a self-hosted chat UI?

The UI itself runs on modest hardware. You need a GPU only if you host models locally at useful speed. Many teams run the UI on a small server and connect it to cloud APIs, adding a GPU box later for private local models.