Deploy and Host self hosted SillyTavern Claude (Self-Hosted Chat and Roleplay Frontend for Claude) on Railway

This template runs SillyTavern wired to Claude, so you can roleplay with Anthropic models on your own server using your own API key.

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

SillyTavern is an open-source LLM frontend, and this template runs the official ghcr.io/sillytavern/sillytavern:1.19.0 image as one service with a single volume at /home/node/app/data.

Why Deploy SillyTavern Claude, the Claude.ai alternative on Railway (Railway Free Trial)

Claude.ai is the easy way in, but you rent a fixed subscription and get no character cards, lorebooks or control over how the prompt is cached.

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 Claude 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 Claude self hosting

Railway runs SillyTavern Claude as one managed service, while a VPS leaves the setup to you. This table compares them.

Provider	What You Get with Railway	What You Get with the Other Provider
DigitalOcean	A template deploy, a public URL and a persistent volume for chats and keys.	A Droplet you size and secure yourself, with Docker and TLS set up by hand.
AWS	One service running the official image, with no VPC or load balancer to configure.	Fine grained networking and scaling, but more pieces before Claude responds.
Hetzner	Managed deploys and a built in domain, so no server upkeep.	Low cost capacity, plus your own OS updates, reverse proxy and restarts.

A VPS can be cheaper per month if you already run servers.

Common Use Cases for hosted SillyTavern Claude

Long roleplay chats with prompt caching: A multi-week story with a big system prompt and a growing history is where caching pays off. Set claude.cachingAtDepth to 0 and most of each turn's input is read from cache.
Character cards on Claude models: Import a card, pick a Claude model in API Connections, and keep the persona and example dialogue as written. Cards from other providers work unchanged.
Group chats: Several characters share one scene and one Claude connection. Caching still helps, as long as the history above the cache depth stays stable.
Lorebook heavy worlds: A world info book that injects many entries per turn grows the prompt fast. Caching the stable part keeps repeat turns cheap.
A shared server with your own keys: Friends or family each log in and paste their own Anthropic key. Keys are stored per user in the data volume, so bills and keys stay separate.

Dependencies for SillyTavern Claude Docker hosted on Railway

This template runs one service, the official SillyTavern 1.19.0 image, with one volume for everything you create.

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

Railway account: a free trial covers a first test, and an idle instance sits near 150 MB of memory.
Anthropic API key: create it in the Anthropic console and paste it into API Connections after login. It's stored per user in the volume secrets, not in a Railway variable.
ADMIN_PASSWORD: generated and re-applied on every boot, so copy it from the Variables tab first.
Persistent volume: keeps chats and config.yaml across redeploys.

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

The image is ghcr.io/sillytavern/sillytavern:1.19.0 with user accounts on, so every API route needs a session.

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

SillyTavern vs Claude.ai (Anthropic's Official Chat App)

Character cards: SillyTavern loads character cards, lorebooks and group chats, which Claude.ai does not offer as a roleplay workflow.
Keys and billing: SillyTavern runs on your own Anthropic API key and bills per token, while Claude.ai is a flat subscription with its own usage limits.
Where Claude.ai wins: It is zero setup, works on any device, and you never touch config files or prompt caching settings.

SillyTavern vs TypingMind (Paid Multi-Model Chat Frontend)

Self hosting: SillyTavern runs on your own Railway service and keeps characters and keys on your volume, while TypingMind is a hosted web app.
Roleplay depth: SillyTavern has lorebooks, presets and group chats built for roleplay, which TypingMind does not focus on.
Where TypingMind wins: Its polished interface, team features and zero maintenance make it easier for people who do not want to run anything.

SillyTavern vs Open WebUI (Self-Hosted ChatGPT-Style UI)

Focus: SillyTavern is built around characters and roleplay, while Open WebUI is a general chat interface centered on local models through Ollama and RAG.
Claude support: SillyTavern exposes Claude-specific options such as prompt caching depth and thinking settings in config.yaml, which Open WebUI does not expose in the same way.
Where Open WebUI wins: It suits a shared team chat server and local model work better than a single roleplay-focused frontend.

SillyTavern Claude vs OpenRouter Claude (Routed Access to Claude Models)

Billing path: SillyTavern Claude sends requests straight to Anthropic with your own key, so you see Anthropic's billing and caching behavior directly.
Caching control: The claude section of config.yaml lets you set enableSystemPromptCache and cachingAtDepth yourself, while OpenRouter applies its own caching rules.
Where OpenRouter wins: One key gives you many models from different providers, which is handy if you want to switch away from Claude often.

How to use SillyTavern Claude (the OSS LLM frontend)?

Open the public URL, click the User account button, and log in as default-user with the ADMIN_PASSWORD value from the Variables tab.
Click the plug icon to open API Connections, then choose Chat Completion as the API and Claude as the Chat Completion Source.
Paste your Anthropic API key and click Connect. The key is saved per user in the data volume secrets, so it survives redeploys.
Pick a Claude model from the model list, then send a test message to confirm the connection works.
Open the Railway service shell and edit config.yaml in /home/node/app/data/_railway. Set the claude section so caching is on:

claude:
  enableSystemPromptCache: true
  cachingAtDepth: 0

Restart the service so the new config loads. Log in again, since the session cookie changes with each deployment.
Chat for a few turns, then check cache reads and writes in the Anthropic console to confirm caching is working and your costs are dropping.

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

Running this on your own VPS works with the upstream Docker image. Replace your-domain with your hostname.

Clone the Repository

Pull the upstream repo to keep your config in one place.

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

Install Dependencies

The official image bundles Node.js and everything SillyTavern needs. Mount a volume so characters, chats and API keys survive upgrades.

docker pull ghcr.io/sillytavern/sillytavern:1.19.0

Configure Environment Variables

Turn on user accounts before exposing the port.

claude:
  enableSystemPromptCache: true
  cachingAtDepth: 0
  extendedTTL: false

Start the SillyTavern Claude Application

Run the container with the port and data volume mapped, open your-domain, log in, and paste your Anthropic key under API Connections.

Official Pricing of SillyTavern Claude (SillyTavern pricing)

SillyTavern itself is free and open source, so there is no SillyTavern subscription to buy.

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

Self hosting SillyTavern Claude means a small always-on container plus your own Anthropic bill, instead of a flat subscription.

Monthly cost of self hosting SillyTavern Claude on Railway

Railway bills by usage, and one always-on SillyTavern instance at light use runs about $5 a month, since it idles near 150 MB.

System Requirements for Hosting SillyTavern Claude on a VPS

SillyTavern is light, idling around 150 MB and peaking near 500 MB during the first-boot compile, so a VPS with 1 GB of RAM handles chat use.

Frequently Asked Questions (FAQs)

Should I use my own Anthropic key or OpenRouter?

Use your own Anthropic key for direct billing and the claude settings in config.yaml.

Why does prompt caching never seem to hit?

Usually something above the cached depth changes between requests, such as a random macro or a timestamp in the system prompt.

Does extendedTTL cost more?

Yes. Check the Anthropic pricing page for current numbers.

Where does config.yaml live on this template?

It lives at data/_railway on the persistent volume, so your edits survive redeploys. Restart the service after editing it so the changes take effect.

What do the thinking settings do?

Reasoning effort controls how much a supported Claude model thinks before answering, and showing thoughts displays that reasoning in chat. enableAdaptiveThinking lets supported models decide when to think, which is a sensible default.

Will Claude refuse some roleplay content?

Anthropic's usage policies apply to the models no matter which frontend you use.