Deploy and Host self hosted SillyTavern Ollama (Self-Hosted SillyTavern on Ollama Models) on Railway

SillyTavern Ollama runs the official SillyTavern image on Railway and connects it to an Ollama service you add to the same project at http://ollama.railway.internal:11434. Railway is CPU-only, so pick small quantized models.

About Hosting SillyTavern Ollama open-source software on Railway (self hosted SillyTavern with Ollama template)

The template runs one service: the official SillyTavern image pinned at 1.19.0, listening on port 8000, with a volume at /home/node/app/data holding characters, chats, lorebooks, presets, keys and the account database. Ollama isn't included, so you add it to the same project.

Why Deploy SillyTavern Ollama, the Open WebUI alternative on Railway (Railway Free Trial)

Open WebUI puts Ollama first, which is fine until you want SillyTavern's character cards, group chats and lorebooks. Deploy this template, add an Ollama service beside it, and chats stay on Railway's private network. The $5 trial covers a light test.

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying SillyTavern Ollama on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.

Railway vs Other Hosting Providers and VPS for SillyTavern Ollama self hosting

Provider	What You Get with Railway	What You Get with the Other Provider
DigitalOcean	Private Ollama on railway.internal plus a data volume, no admin. CPU-only, small models.	A Droplet you secure and patch, with a volume for models.
AWS	Template setup with volumes for chats and models. No GPUs, so replies are slow.	EC2 GPU boxes for real speed, but IAM and EBS are on you.
Hetzner	Same private network with less setup, though Hetzner CPU can outrun a Railway container.	Cheap CPU and storage; you own the firewall and backups.

Common Use Cases for hosted SillyTavern Ollama

Private character chats: Ollama runs in your Railway project, so prompts never reach OpenAI or OpenRouter. Cards stay between you and your model.
Trying roleplay models quickly: pull a 3B to 8B quantized model, chat, pull another. One volume holds them and API Connections switches with a dropdown.
A small always-on model for testing cards: one modest model loaded is enough to check greetings, example dialogue and lorebook entries. Slow on CPU, fine for a few messages.
Chats and models in one project: characters and lorebooks sit on SillyTavern's volume, weights on Ollama's, both on the private network. One bill.
Switching APIs per chat: Ollama for private scenes, a hosted provider when you want fast long replies. Save each as a connection profile and switch in API Connections.

Dependencies for SillyTavern Ollama Docker hosted on Railway

The template ships one service: SillyTavern on ghcr.io/sillytavern/sillytavern:1.19.0, on port 8000 with a volume at /home/node/app/data. Ollama is not included.

Deployment Dependencies for Managed SillyTavern Ollama Service (LLM Chat Frontend)

Settings are SILLYTAVERN_ variables overriding config.yaml keys: SILLYTAVERN_PORT=8000, SILLYTAVERN_LISTEN=true, SILLYTAVERN_WHITELISTMODE=false, SILLYTAVERN_ENABLEUSERACCOUNTS=true, PUID=1000, PGID=1000, and ADMIN_PASSWORD.

Implementation Details for SillyTavern Ollama (Using SillyTavern official docker image)

The image is pinned and the volume holds characters, lorebooks, secrets and accounts; idle memory is about 150 MB. Railway has no GPUs, so pull 3B to 8B quantized models and expect slow CPU replies.

How does SillyTavern Ollama compare against other ways to chat with Ollama models

SillyTavern vs Open WebUI (Open WebUI Alternative)

SillyTavern wins: Character cards, group chats and lorebooks run deeper than Open WebUI's document chats.
Open WebUI wins: Built-in RAG with citations and shared model settings suit a team server.
Wiring: Both hit http://ollama.railway.internal:11434, but SillyTavern needs the Ollama API type picked by hand.

SillyTavern vs KoboldCpp Lite (KoboldCpp Lite Alternative)

SillyTavern wins: You swap Ollama models by reconnecting, without restarting the frontend process.
KoboldCpp wins: One process is the model and the UI, with samplers built in, so there's no second service.
Railway fit: Separate services restart independently, handy when a pull fails. CPU inference is slow either way.

SillyTavern vs Agnaistic (Agnaistic Alternative)

SillyTavern wins: More extensions, image generation, TTS and native Ollama support, plus a bigger card ecosystem.
Agnaistic wins: Shared multi-user roleplay is simpler out of the box if that's all you need.
Wiring: SillyTavern's native Ollama API type is a dropdown plus a URL, with model listing built in.

SillyTavern vs LM Studio (LM Studio Alternative)

SillyTavern wins: It stays always-on at your Railway domain so friends log in; LM Studio is a desktop app on your own machine.
LM Studio wins: GPU acceleration and a model browser make local testing faster and simpler.
Middle ground: Point SillyTavern at LM Studio's server locally for speed, or keep Ollama private on Railway for uptime.

How to use SillyTavern Ollama (the OSS LLM chat frontend)?

Add an ollama/ollama service to the same Railway project, mount a volume at /root/.ollama, set OLLAMA_HOST=[::]:11434, and keep it private.
Open a Railway shell on it, pull a quantized model, and confirm:
```
ollama pull llama3.2:3b
ollama list
```
Log in to SillyTavern on your Railway domain with ADMIN_PASSWORD, then open API Connections from the plug icon.
Choose Text Completion, API type Ollama, server URL http://ollama.railway.internal:11434, then Connect.
Select the model; in Advanced Formatting set context size and the instruct template the model recommends, and raise num_ctx to match.
Load a character and chat. CPU inference is slow, so start at 3B and expect pauses.

How to self host SillyTavern Ollama on other VPS Services (SillyTavern self hosting guide)

On a VPS you own the config, the process and the upgrades. Docker with volumes for data, config and plugins works, or clone and run Node.js. Ollama installs separately on port 11434.

Clone the Repository

Pull a tagged release rather than staging.

git clone https://github.com/SillyTavern/SillyTavern.git
cd SillyTavern
git checkout 1.19.0

Install Dependencies

Use a Node.js LTS release, then install:

npm install

The Docker image skips this step.

Configure Environment Variables

Back up config.yaml first. Set listen to true, then pick user accounts or the IP whitelist; behind a proxy the whitelist blocks everyone, so use accounts with HTTPS.

Start the SillyTavern Ollama Application

Run node server.js behind nginx or Caddy for TLS, log in on your-domain, and add http://127.0.0.1:11434 as the Ollama URL in API Connections.

Railway handles the proxy, TLS, restarts, volume and health checks for you.

Official Pricing of SillyTavern Ollama (SillyTavern pricing)

SillyTavern itself is free and open source under AGPL-3.0. There is no license fee and no per-seat charge, so the pricing question is really about infrastructure.

SillyTavern Ollama cloud vs self hosted comparison (Pricing, features, costs, and more)

Self hosting keeps the chat frontend and the model server in one project you control, but the cost picture differs from a hosted character chat subscription.

Monthly cost of self hosting SillyTavern Ollama on Railway

Budget about $5/mo for SillyTavern itself, since it idles near 150 MB; the Ollama service is the real cost, because a 3B to 8B model keeps several GB of RAM busy while loaded. The $5 trial covers the first stretch of testing.

System Requirements for Hosting SillyTavern Ollama on a VPS

You need Docker, a volume for SillyTavern data, and enough RAM for the quantized models you pull. A 3B to 8B model wants a few gigabytes free alongside the frontend. Without a GPU, replies come slowly, so size the box for patience rather than throughput.

Frequently Asked Questions (FAQs)

Text Completion or Chat Completion with Ollama?

Text Completion with API type Ollama is the documented path and gives you instruct template control in Advanced Formatting. Chat Completion through Ollama's OpenAI-compatible /v1 endpoint works too, and lets Ollama apply the model's chat template itself.

Why does the Ollama connection fail?

Usually the server URL or the Ollama service itself. Use http://ollama.railway.internal:11434, confirm the service runs with OLLAMA_HOST=[::]:11434, and pull a model first; SillyTavern can only list what Ollama already has.

How slow is CPU inference on Railway?

Railway has no GPUs. A small quantized 3B to 8B model answers in tens of seconds to a few minutes depending on prompt size and context, so keep context modest and expect to wait.

Which model sizes are practical?

3B to 8B quantized builds are the realistic range. Larger models load slowly or not at all in the memory a Railway instance gives you.

Why does it ramble or speak for my character?

That's the instruct template and stop strings, not Ollama. Pick the template recommended for your model in Advanced Formatting, set context size to match num_ctx, and trim the prompt.

Does Ollama need a public domain?

No. Keep it private and reach it over railway.internal. If you enable SILLYTAVERN_PRIVATEADDRESSWHITELIST_ENABLED=true, allow 10.0.0.0/8 and fd00::/8 so SillyTavern isn't blocked.