Run a private ChatGPT-style chat on Railway

By Tracklayer · Updated 2026-11-20

Short answer: deploy a self-hosted chat interface such as Open WebUI on Railway. Tracklayer’s Open WebUI template runs Open WebUI v0.12.0 with Ollama for open models, SearXNG for private web search, Postgres with pgvector for document chat, and Redis. The admin account is created at deploy and public sign-up is closed. Railway has no GPUs, so keep local models small or add an API key for hosted ones.

Deploy Open WebUI with Ollama on Railway

What “private” means here

A ChatGPT-style interface lets a team talk to language models through a browser. Self-hosting it means your conversations, uploaded documents and user accounts live in a database you control, instead of in a vendor’s workspace. Uses named in the templates:

If you send prompts to a hosted provider such as OpenAI, those prompts go to that provider. Only models running in Ollama keep prompts on your Railway project.

Which chat UI templates people deploy on Railway

TemplateCreatorDeploymentsServices
TypebotBrody's Projects8,9687
LibreChatLibreChat6,7625
Typebot (Atualizado)Plugr.dev (Verified)2,6075
Open WebUIalphasec1,0981
Lobe ChatLobeHub6771
DialoqbaseMuhammed Nazeem's Projects6033
Open WebUI with PipelinesTom5332
SillyTavernnull4071

Railway marketplace data as of 2026-11-19. Full ranking: Best self-hosted ChatGPT-style UIs on Railway.

Check each template’s services before choosing: some include a model server, a vector database or search, and some are only the interface.

Deploy Open WebUI with Ollama, step by step

These steps use Tracklayer’s Open WebUI with Ollama template.

  1. Enter your email. Put it in WEBUI_ADMIN_EMAIL when deploying.
  2. Sign in. After the first deploy, copy WEBUI_ADMIN_PASSWORD from the Open WebUI service variables and sign in at the public URL. Change the password under Settings, Account.
  3. Wait for the models. Ollama pulls llama3.2:3b and nomic-embed-text:v1.5 on first boot, about 2.3 GB. Start chatting once its logs show both pulled.
  4. Add users. Use Admin Panel, Users. New sign-ups stay pending until approved.
  5. Add models. Pull more from Admin Settings, Models, or add an OpenAI-compatible key under Connections for larger hosted models.

Open WebUI saves most settings in its database after the first start. Later changes belong in Admin Settings, because editing the matching variables has no effect unless ENABLE_PERSISTENT_CONFIG=false.

What is stored where

Open WebUI keeps uploads and caches on a volume. Ollama keeps models on its own volume. Users, chats, settings and RAG vectors live in Postgres with pgvector. Nothing is public except the Open WebUI service.

Costs

Railway bills usage. In this stack Ollama is the main driver, because model memory is roughly the model file size plus context: the template puts a 3B model at about 3 GB and a 7 to 8B model at 6 to 8 GB.

The template’s resource table estimates about 0.7 GB of RAM idle and about 4.5 GB under typical use, with about 5.2 GB of volume. These are estimates, not measurements. It lists Hobby as the minimum plan, because the 1 GB per-service limit on the Trial plan is too small for a 3B model.

A lighter option: LibreChat

If you mainly want one interface for hosted providers such as OpenAI, Anthropic, Google and OpenRouter, Tracklayer’s LibreChat template deploys LibreChat v0.8.8 with MongoDB, Meilisearch and a RAG API on Postgres with pgvector. It does not bundle a model server.

Limits and licence

Other options

See the full ranking of ChatGPT-style UIs on Railway. To build apps and RAG pipelines rather than chat, see AI app builder templates. For all guides, see /guides/.

FAQ

Can it run large models on Railway?

Railway has no GPUs, so Ollama runs models on the CPU and the Open WebUI template recommends keeping them small. For larger models, add an OpenAI-compatible API key under Admin Settings, Connections.

How do I log in to the Open WebUI template?

Enter your email in WEBUI_ADMIN_EMAIL when deploying. After the first deploy, copy WEBUI_ADMIN_PASSWORD from the Open WebUI service variables and sign in at the public URL. Public sign-up is closed, and new sign-ups stay pending until an admin approves them.

What does a private chat setup cost?

For Open WebUI with Ollama, the template estimates about 4.5 GB of RAM under typical use, mostly Ollama, and lists Hobby as the minimum plan. For LibreChat, the template measured about 1.3 GB of RAM idle across five services and estimates $15 to $30 per month for a small team, plus model API costs.

Are there licence terms to know?

Open WebUI uses its own BSD-3-based licence with a branding clause: its name and logo must stay visible unless you have 50 or fewer end users in a 30-day period, written permission, or an enterprise licence. LibreChat's source repository is listed as MIT in the template's NOTICE file.