Self-hosted ChatGPT alternatives: Open WebUI vs LibreChat vs big-AGI
Compare three open-source ChatGPT-style interfaces you can self-host on Railway, where each one stores your chats, and which to pick for you or your team.By chin · Oct 4, 2026 · 4 min read
A hosted chatbot is the fastest way to try AI. It is also a place where your prompts, files and API spend live on someone else's terms. A self-hosted chat interface flips that around. You bring your own API keys (OpenAI, Anthropic, Groq, OpenRouter or a local model), you choose who can sign in, and conversation history stays in a database you control.
This guide compares the three open-source interfaces available as one-click Railway templates on this site. It covers what each is best at, where each keeps your data, and the setup details that matter in production.
The short answer
| Best for | Chats stored in | Model keys | |
|---|---|---|---|
| Open WebUI | Teams wanting a ChatGPT-like app | Server (add a volume) | Admin panel |
| LibreChat | Many providers, searchable history | Bundled MongoDB | Variables |
| big-AGI | Comparing models side by side | Your browser | In the app |
None of the three asks for a key at deploy time, so you can deploy first and connect models afterwards.
If you only want the simplest possible "ChatGPT with my own key", the ChatGPT-webui template asks for a single
OPENAI_API_KEY and nothing else.Open WebUI: the closest thing to ChatGPT
Open WebUI is the most popular self-hosted chat interface, and it looks and feels a lot like ChatGPT. It supports multiple users with roles, model presets, document uploads and any OpenAI-compatible API. That includes OpenRouter, Groq, and local models served through Ollama.
The Open WebUI with Artifacts template is built from a fork that improves code artifacts, so generated HTML, React or SVG renders in a side panel, similar to v0 or Claude's artifacts.
After deploying:
- Open the generated domain and create the first account. The first user becomes the admin.
- Go to Admin Panel → Settings → Connections and add your OpenAI-compatible endpoint and key.
- If the URL is public, go to Admin Panel → Settings → General and disable new sign-ups, or set new users to "pending" so you approve each one.
Persist your data. This template does not attach a volume, so users, chats and settings are lost when the service is redeployed. Before you invite anyone, attach a Railway volume to the service mounted at
/app/backend/data, which is where Open WebUI keeps its database and uploads.LibreChat: multi-provider, with searchable history
LibreChat is built for using many model providers from one interface. You can switch between OpenAI, Anthropic, Google, Groq and others mid-conversation. Each provider is enabled with a variable, and setting a key to
user_provided lets every user bring their own key instead of sharing yours.The LibreChat Lite template deploys three services:
- LibreChat, the app itself
- MongoDB, which stores users and conversations on a volume
- Meilisearch, which powers full-text search across your chat history
"Lite" means it ships without LibreChat's separate RAG API and vector database. That keeps it cheap to run. The trade-off is that chatting with large document collections is out of scope.
After deploying:
- Add provider keys in the LibreChat service's Variables tab (for example
GROQ_API_KEYorGOOGLE_KEY), then redeploy. - Register your own account, then set
ALLOW_REGISTRATION=falseso strangers who find the URL cannot sign up and spend your credits.
big-AGI: for comparing models
big-AGI takes a different approach: it is a single lightweight service, and your chats and API keys are stored in your browser, not on the server. That makes the big-AGI template the easiest to run and the cheapest to host. It also means history does not sync between devices unless you export it.
Its standout feature is Beam. It sends the same prompt to several models at once and lets you merge the best parts of each answer. That is useful when you are choosing a model for a project, or when you want a second opinion on important output.
Because keys are stored client-side, big-AGI is best for personal use. For a team, Open WebUI or LibreChat gives you real accounts and central control.
Which one should you pick?
- Just you, one provider, minimum effort: ChatGPT-webui with your
OPENAI_API_KEY. - Just you, many models, lots of experimenting: big-AGI.
- A small team that wants a familiar ChatGPT experience: Open WebUI, with a volume attached.
- A team using several providers, or letting users bring their own keys: LibreChat.
Keeping costs predictable
Two costs add up: the Railway resources for the app (and database, if bundled), and the per-token cost of the models you call. To keep both in check:
- Set a usage limit in your Railway workspace so a runaway service cannot surprise you.
- Use provider-side spending caps on your API keys.
- Prefer smaller, faster models (Groq-hosted open models, or the "mini" tier of the big providers) for everyday chats, and save the frontier models for hard problems.
Frequently asked questions
Can I use these with local models instead of paid APIs?
Yes. Open WebUI and LibreChat both work with any OpenAI-compatible endpoint, including Ollama. Running the model itself on Railway needs a lot of memory, so most people point these interfaces at a hosted API or a GPU machine they already have.
Is my data private?
Your conversation history stays in your own deployment. Prompts are still sent to whichever model provider you configure, so check that provider's data-retention policy.
Do I need to know Docker?
No. Each template comes pre-configured: click Deploy, fill in any requested keys, and Railway builds and runs everything.
Templates in this guide
More guides
- Build a self-hosted Perplexity alternative with Morphic, SearXNG and FirecrawlSelf-host an AI answer engine with cited sources. Compare Morphic, Open DeepResearch and Fireplexity, and add private search with SearXNG or Whoogle.
- How to self-host Excalidraw with real-time collaborationRun your own Excalidraw whiteboard with live, end-to-end encrypted collaboration on Railway. Covers how the two services fit together, setup, and common fixes.
- Self-hosted PDF tools: Stirling-PDF vs PDFDing vs Paperless-AIEdit, organise and auto-tag PDFs without uploading them to online converters. A practical comparison of three self-hosted PDF tools you can deploy on Railway.