How to Set Up SillyTavern in the UK: A Beginner's Guide

SillyTavern gets recommended constantly on Reddit as the most private, flexible Character AI alternative — and then people open the GitHub repo, see the word 'Node.js', and quietly close the tab. It has a reputation for being fiddly, and for UK users there's the extra wrinkle of working out which API backends actually accept UK cards, comply with the way HMRC-registered services handle VAT, and don't throttle you into oblivion at 8pm on a Sunday. This guide walks through the whole process from a British beginner's perspective: what SillyTavern actually is (a frontend, not an AI), what you need to install on Windows, macOS or Linux, and how to hook it up to a language model using OpenRouter or a locally-run model on your own PC. You'll also learn the sensible defaults for context length, sampler settings and character cards so your first roleplay session doesn't feel like a broken chatbot from 2018. By the end you'll have a working setup that costs pennies per session, keeps your chats on your own machine, and gives you far more control than any hosted platform in the UK market.

What SillyTavern actually is (and isn't)

The single biggest source of confusion for new users is thinking SillyTavern is an AI. It isn't. SillyTavern is a frontend — a browser-based interface that runs on your own computer and talks to a separate language model somewhere else. That 'somewhere else' can be a cloud API like OpenAI, Anthropic or OpenRouter, or a model you're running locally on your own hardware using something like KoboldCpp or Oobabooga's text-generation-webui.

Think of it like Thunderbird for email. Thunderbird doesn't host your inbox — Gmail or Outlook does. Thunderbird just gives you a nicer way to read and organise it. SillyTavern is the same idea for AI roleplay: it gives you character cards, chat history management, group chats, world info (lorebooks), regex extensions, text-to-speech, image generation hooks and dozens of other features that hosted platforms simply don't offer. But it needs a model to actually generate words.

This architecture is why it consistently tops privacy comparisons. Your character cards, chat logs and settings live in a folder on your PC. Nobody's data-mining your conversations to train the next model, and there's no terms-of-service update that suddenly bans your favourite scenario. The tradeoff is that you're now responsible for the plumbing — which is what the rest of this guide covers.

It's worth being honest about who SillyTavern suits. If you want to open an app, tap a character and start chatting in under thirty seconds, look at hosted UK-accessible options instead. SillyTavern is aimed at people who enjoy tinkering, want deep customisation, or specifically need the privacy of self-hosting. If that's not you, that's fine — there's no shame in preferring a polished mobile app.

What you need before installing

SillyTavern runs on Windows 10 or 11, macOS (Intel or Apple Silicon) and most mainstream Linux distributions. It's lightweight — the frontend itself uses barely any resources, since all the heavy lifting happens on the API side or on whatever separate model backend you're running.

You'll need three things installed before you start. First, Node.js version 18 or later, which is the JavaScript runtime SillyTavern is built on. Grab the LTS installer from nodejs.org and accept the defaults. Second, Git, which is used to download and update SillyTavern cleanly. Windows users can get Git for Windows from git-scm.com; macOS users usually already have it via Xcode command line tools; Linux users can install it through their package manager. Third, a decent modern browser — Firefox, Chrome or Edge are all fine.

Disk space isn't a concern for SillyTavern itself (under 500MB), but if you plan to run a local model you'll want a machine with a reasonable GPU. An Nvidia card with 8GB or more of VRAM will comfortably run a good 7B or 8B model at decent speeds. Apple Silicon Macs (M1 and later) also handle local models well thanks to unified memory. If your hardware is older or you're on a laptop with integrated graphics, stick to a cloud API — it'll be cheaper and faster than fighting with a local setup.

One UK-specific note: if you're on a rural broadband connection with tight data caps, be aware that downloading a local model can pull 5-20GB depending on quantisation. The SillyTavern install itself is small, but the model files aren't.

• Node.js 18 or later (LTS recommended)
• Git
• A modern browser
• 8GB+ VRAM if running models locally, otherwise a cloud API

Installing SillyTavern step by step

Open a terminal — on Windows that's PowerShell or Windows Terminal, on macOS it's Terminal.app, on Linux whatever you normally use. Navigate to wherever you want SillyTavern to live. A folder in your Documents or user directory is fine; avoid Program Files on Windows because it causes permission headaches later.

Run this command to clone the repository:

git clone https://github.com/SillyTavern/SillyTavern -b release

The '-b release' bit is important — it grabs the stable branch rather than the development one, which occasionally breaks. Once the download finishes, change into the new folder with 'cd SillyTavern'.

On Windows, double-click Start.bat in File Explorer, or run it from the terminal. On macOS and Linux, run './start.sh' (you may need to make it executable first with 'chmod +x start.sh'). The first launch takes a minute or two while npm downloads dependencies. When it's done, your browser should automatically open to http://localhost:8000 with the SillyTavern interface loaded.

If the browser doesn't open, just visit that address manually. If you see an error about a port being in use, another program is squatting on 8000 — edit config.yaml in the SillyTavern folder and change the listen port to something else like 8001.

Updating later is just as painless: close SillyTavern, open the folder in a terminal and run 'git pull'. Then start it again. The release branch updates roughly once a month with new features and bug fixes. Your character cards, chats and settings live in the /data folder and are preserved across updates, but it's still worth copying that folder somewhere safe occasionally as a backup.

Choosing a UK-accessible API backend

This is where UK users have historically hit walls. Some backends require US phone verification, some don't accept prepaid or Monzo/Starling cards, and some geo-block certain content. Here's what actually works well from the UK in 2024 and 2025.

OpenRouter is the pragmatic default. It's a marketplace that gives you a single API key and access to dozens of models from Anthropic, OpenAI, Meta, Mistral, DeepSeek and various fine-tunes tuned for roleplay. It accepts standard UK debit and credit cards, charges in USD (your bank handles the conversion), and lets you top up in small amounts rather than committing to a monthly subscription. For SillyTavern, the free-tier and low-cost models are more than enough to get started — you can spend under a pound and get hours of chatting.

OpenAI direct works from the UK without issue if you want GPT-4-class quality, but their content policies are stricter and they're expensive for long roleplay sessions. Anthropic's Claude API is similar — excellent prose, tight moderation. For unfiltered creative writing, many UK users route through OpenRouter to specific community models rather than going direct.

Local models are the third path. Download KoboldCpp (a single executable), grab a GGUF-format model from Hugging Face — something like a Llama 3 8B or Mistral Nemo 12B finetune is a good starting point — and point SillyTavern at KoboldCpp's local address. Zero ongoing cost, complete privacy, but limited by your hardware.

In SillyTavern, click the plug icon at the top, choose 'Chat Completion' or 'Text Completion' depending on your backend, select the source (OpenRouter, OpenAI, KoboldCpp, etc.), paste your API key or set the local URL, and pick a model from the dropdown. Save. That's the connection done.

Loading your first character and configuring settings

Character cards are the personality files SillyTavern uses. They're PNG images with the character's description, greeting, example dialogue and behaviour baked into the file's metadata. The community hub Chub.ai is the biggest public library; you can also make your own inside SillyTavern using the character editor.

Download a card as a PNG, then in SillyTavern click the user icon on the right, hit the import button, and select the file. The character appears in your list. Click them, and the greeting message loads into a fresh chat.

Before you start chatting, tweak two things. First, the context size — this is how much conversation history the model 'remembers'. In the sampler settings (the sliders icon), match this to your model's capability. OpenRouter models typically support 8k to 128k tokens; local 7B models usually 8k. Setting it too high wastes money on cloud APIs and slows local models; too low and the character forgets what happened three messages ago.

Second, the temperature. This controls randomness. Around 0.7 to 0.9 is the sweet spot for coherent but creative roleplay. Below 0.5 gets repetitive; above 1.2 gets incoherent. Every model has a slightly different ideal, but those are safe starting points.

Instruction templates matter too — if your model expects ChatML format and SillyTavern sends Alpaca, quality drops sharply. For OpenRouter Chat Completion models this is handled automatically. For local models via Text Completion, check the model's Hugging Face page for the correct template and select it in the AI Response Formatting menu.

If you're coming from a hosted service and want a smoother comparison point while you learn, our guide to Character AI alternatives covers the trade-offs between hosted and self-hosted options in more depth.

Common UK-specific gotchas and troubleshooting

A few issues come up repeatedly for British users. If OpenRouter rejects your card, check whether it's a virtual or prepaid card — some issuers block cross-border USD transactions by default. A quick call to your bank or toggling international payments on in the app usually fixes it. Monzo, Starling, Revolut and standard high street bank cards all generally work.

If SillyTavern won't start and the terminal shows 'node is not recognised', Node.js either isn't installed or isn't on your system PATH. Reinstall from nodejs.org and tick the option to add it to PATH during setup, then restart your terminal.

Slow responses from cloud APIs are usually a model issue, not a location issue. Latency from the UK to US-based OpenRouter endpoints is typically 100-200ms, which is invisible during chat. If generation itself feels sluggish, you've picked a large model — swap to a faster one.

For local models on Windows, if KoboldCpp crashes on launch, you likely downloaded a model too large for your VRAM. GGUF files come in different quantisations (Q4_K_M is a good balance); check the file size against your available VRAM plus a bit of headroom.

Finally, on privacy: SillyTavern itself doesn't phone home, but any cloud API you connect to obviously sees your prompts. If that concerns you, run local. If you use OpenRouter, note that individual providers on the network have their own logging policies — the OpenRouter dashboard lets you filter to providers that don't log.

Frequently asked