SillyTavern Complete Guide 2026: Setup, Models, and Tips
SillyTavern ruined every other AI chat interface for me. once you've used it, going back to the basic ChatGPT web UI feels like going from a sports car to a bicycle. it's free, runs locally, supports every major API, and gives you more customization than any commercial alternative.
tl;dr: SillyTavern is a free, locally-installed AI chat frontend that supports every major API. connect it to NanoGPT for 50+ models at $8/month, or use Ollama for completely private local AI. setup takes 15 minutes with Node.js.
key takeaways:
- SillyTavern runs on your machine so your data stays local, regardless of which API you connect to
- connect to NanoGPT for the best value: one API key gives you Claude 3.5 for writing, GPT-4o for code, and 50+ other models
- the Summarize and Vector Storage extensions are essential for keeping context in long conversations
i've been running SillyTavern daily for months - for creative writing, roleplay, and even coding assistance. here's everything i know.
👉 Connect SillyTavern to NanoGPT – access 50+ models for $8/month instead of paying for each API separately
what is SillyTavern?
short answer: SillyTavern is a locally-installed web app that acts as a chat frontend for AI models. it runs on your machine, supports any API, and has a character system for roleplay.
SillyTavern is a locally-installed web application that acts as a chat frontend for large language models. think of it as a supercharged chat UI that you control completely.
key things:
- runs on your machine - your data stays local
- connects to any API - OpenAI, Anthropic, NanoGPT, Ollama, KoboldAI, and more
- character system - create and share AI characters with custom personalities
- extensions - add features like image generation, text-to-speech, web search
- advanced prompting - fine-tune system prompts, temperature, samplers
it's built on Node.js and runs in your browser once started. available on Windows, Mac, and Linux.
installing SillyTavern
requirements
- Node.js 18+ (LTS recommended)
- Git
- 2GB free disk space
- any modern browser
Windows installation
# install Node.js from nodejs.org first
# then open PowerShell:
git clone https://github.com/SillyTavern/SillyTavern.git
cd SillyTavern
.\start.bat
Mac/Linux installation
git clone https://github.com/SillyTavern/SillyTavern.git
cd SillyTavern
chmod +x start.sh
./start.sh
first launch takes a minute to install dependencies. after that, it opens in your browser at http://localhost:8000.
update SillyTavern
cd SillyTavern
git pull
# restart the application
update regularly - new features and bug fixes come out frequently.
connecting SillyTavern to AI APIs
short answer: connect to NanoGPT (recommended for value), OpenAI direct, Anthropic direct, or Ollama for local private AI.
this is the important part. SillyTavern supports multiple API backends. here's how to connect each one:
option 1: NanoGPT (recommended)
best value for money. one API key, 50+ models.
- get your API key from NanoGPT - see our API key setup guide
- in SillyTavern: click the API connection icon (plug icon, top left)
- select "Chat Completion" as API type
- choose "OpenAI" as the provider
- paste your NanoGPT API key
- set the endpoint to:
https://api.nano-gpt.com/v1 - click "Connect"
- select a model from the dropdown
our dedicated SillyTavern + NanoGPT setup guide has screenshots for every step. for maximum privacy, see our private API setup.
option 2: OpenAI direct
- get an API key from platform.openai.com
- in SillyTavern: API type → Chat Completion → OpenAI
- paste your key
- connect and select model
option 3: Anthropic direct
- get an API key from console.anthropic.com
- API type → Chat Completion → Claude
- paste key, connect
option 4: Ollama (local)
for completely private, offline use:
- install Ollama from ollama.ai
- pull a model:
ollama pull llama3:8b - in SillyTavern: API type → Chat Completion → Ollama
- endpoint:
http://localhost:11434 - connect
API comparison
| API | Models | Cost | Privacy | Best For |
|---|---|---|---|---|
| NanoGPT | 50+ | From $8/mo | Good | Daily use, variety |
| OpenAI | 3-4 | Pay-per-use | Poor | GPT-4o only |
| Anthropic | 2-3 | Pay-per-use | OK | Claude only |
| Ollama | Many | Free | Excellent | Sensitive data |
best models for SillyTavern
after testing 20+ models in SillyTavern over 2 months, here are my recommendations:
short answer: Claude 3.5 Sonnet for creative writing and roleplay, GPT-4o for coding, and Mistral Large for budget use.
for creative writing / roleplay
| Model | Quality | Speed | Cost | Notes |
|---|---|---|---|---|
| Claude 3.5 Sonnet | ⭐⭐⭐⭐⭐ | Fast | Medium | best prose quality |
| GPT-4o | ⭐⭐⭐⭐ | Fast | Medium | great all-rounder |
| Claude 3 Haiku | ⭐⭐⭐ | Very Fast | Low | good for quick exchanges |
| Mistral Large | ⭐⭐⭐⭐ | Fast | Low | surprising quality |
my pick: Claude 3.5 Sonnet. the writing quality is noticeably better than GPT-4o for creative tasks. it handles character voice, emotional nuance, and long-form narrative better.
for coding assistance
GPT-4o wins here. use it in SillyTavern when you want a conversational coding partner.
for budget use
Mistral Large or GPT-4o-mini through NanoGPT. cheap enough that you can run long conversations without worrying about cost.
see our detailed guides: best models for roleplay, best models for writing, best models for coding.
essential SillyTavern extensions
extensions make SillyTavern go from good to incredible. here are the ones i actually use:
short answer: Summarize (auto-summarizes long conversations), Vector Storage (lets AI remember past conversations), TTS, and Image Generation.
must-have extensions
- Summarize - auto-summarizes long conversations so the model doesn't forget context
- Vector Storage - lets the AI "remember" things from past conversations
- TTS (Text-to-Speech) - reads AI responses aloud with various voice options
- Image Generation - connects to Stable Diffusion or DALL-E for inline images
nice-to-have extensions
- Web Search - lets the AI search the internet during conversations
- Translate - auto-translate between languages
- Regex Scripts - customize how text is displayed and processed
- Character Expressions - shows character portrait emotions
installing extensions
- in SillyTavern: menu → Extensions → Install Extension
- paste the GitHub URL of the extension
- restart SillyTavern
- enable the extension in settings
SillyTavern settings for best results
these settings made the biggest difference in my testing:
sampler settings
| Setting | Recommended Value | Notes |
|---|---|---|
| Temperature | 0.7-1.0 | higher = more creative, lower = more consistent |
| Top-P | 0.9 | default is fine for most cases |
| Top-K | 20-40 | limits token choices |
| Min-P | 0.05-0.1 | newer sampler, works well |
| Repetition Penalty | 1.05-1.15 | prevents repetitive output |
context settings
- Context size: set to model maximum (8K for GPT-4o, 200K for Claude 3.5)
- Response length: 300-800 tokens for chat, 1000+ for writing
- System prompt: this is where you define the AI's personality - spend time here
prompt formatting
different APIs expect different prompt formats:
- OpenAI / NanoGPT: use "ChatML" format
- Anthropic: use "Assistant" format
- Ollama: depends on the model, usually "ChatML" or "Alpaca"
SillyTavern auto-detects most of the time. if you get weird responses, check the prompt format.
SillyTavern vs alternatives
| Feature | SillyTavern | Open WebUI | LibreChat | ChatGPT |
|---|---|---|---|---|
| Price | Free | Free | Free | $20/mo |
short answer: SillyTavern wins for deep customization and creative tools. Open WebUI is simpler for basic chat. LibreChat is better for teams.
| Self-hosted | Yes | Yes | Yes | No | | Character System | Yes | No | No | No | | Multiple APIs | Yes | Yes | Yes | No | | Extensions | Yes | Limited | Limited | Plugins | | Privacy | Excellent | Good | Good | Poor | | Ease of Use | Medium | Easy | Easy | Easy | | Best For | Creative/Roleplay | General use | Teams | Casual use |
Open WebUI is simpler and better for basic chat. see our Open WebUI setup. LibreChat is better for teams. LibreChat setup.
SillyTavern wins when you want deep customization, character management, or creative writing tools.
troubleshooting common issues
"Connection Failed"
- check your API key is correct
- verify the endpoint URL
- make sure your API account has credits
short answer: check your API key, verify the endpoint URL, make sure your account has credits, and try regenerating the key if connection fails.
- try regenerating the API key
"Context Length Exceeded"
- reduce context size in settings
- enable the Summarize extension
- start a new conversation
Slow Responses
- switch to a faster model (GPT-4o-mini, Claude Haiku)
- check your internet connection
- reduce max response length
Weird/Garbage Output
- check prompt format matches your API
- lower temperature if output is too random
- reset sampler settings to defaults
- try a different model
Extensions Not Working
- update SillyTavern (
git pull) - check extension compatibility with your version
- restart SillyTavern after installing
next steps
- install SillyTavern (15 minutes)
- connect to NanoGPT (setup guide)
- pick a model (Claude 3.5 Sonnet for writing, GPT-4o for code)
- install the Summarize extension
- create your first character
- tweak sampler settings as you go
it's a rabbit hole in the best way. once you get SillyTavern dialed in, you'll never go back to the basic ChatGPT interface.
👉 Get NanoGPT for SillyTavern – 50+ models, $8/month, works perfectly with SillyTavern
Last updated: July 2026
Related Articles
- SillyTavern + NanoGPT Setup - step-by-step connection guide
- Private SillyTavern API - privacy-focused setup
- Best Models for Roleplay - model recommendations
- Local LLM Guide - run models locally with Ollama
- NanoGPT Complete Guide - everything about NanoGPT
- AI Privacy Guide - protect your data while using AI
Disclosure: this article contains affiliate links. if you sign up through our links, we earn a small commission at no extra cost to you. we only recommend tools we personally use and pay for.