SillyTavern is a free, open-source chat interface for people who want more control than ChatGPT-style apps give you. It does not run AI models itself.
Instead, it connects to a model running on your PC (through Ollama, KoboldCpp or llama.cpp) or to a cloud API (OpenAI, Claude, Gemini, OpenRouter and more), and wraps it in a powerful front end with custom characters, lorebooks, group chats, image generation, voices and extensions.
People use it for interactive storytelling, writing practice, language learning with a patient conversation partner, game masters for tabletop sessions, and testing how different models handle the same prompt.
This guide covers installing SillyTavern on Windows, Mac, Linux and Android, connecting a local model, creating your first character, the key settings to change, and fixes for the most common problems.
Details come from the official SillyTavern docs and repo (version 1.19.0), checked in November 2026.
- What: free AGPL-3.0 “LLM frontend for power users”. It needs a separate model back end.
- Needs: Node.js 20 or newer (latest LTS recommended) and Git.
- Install:
git clone https://github.com/SillyTavern/SillyTavern -b release, then double-clickStart.bat(Windows) or run./start.sh. - Opens at:
http://127.0.0.1:8000 - Easiest local back end: Ollama (official docs call it the easiest llama.cpp-based option). KoboldCpp is also popular.
- Update:
UpdateAndStart.baton Windows, orgit pull. Your data lives indata/default-user. - Do not install in Program Files or run Start.bat as admin.
What Is SillyTavern?
SillyTavern (often shortened to ST) is a locally installed web app that runs a small Node.js server on your computer and opens in your browser. Think of it as a cockpit for AI chat: you bring the engine (a model), and ST gives you the controls.
| Feature | What it does |
|---|---|
| Characters | Reusable personas with a description, personality, scenario and example dialogue |
| World Info (lorebooks) | Facts that are added to the prompt only when keywords come up, saving context |
| Group chats | Several characters in one conversation |
| Prompt control | Edit every part of the prompt, sampler settings and instruct templates |
| Extensions | Image generation, text-to-speech, speech input, translation, summarisation, Data Bank (RAG) and more |
| Any back end | Local (Ollama, KoboldCpp, llama.cpp, text-generation-webui, TabbyAPI) or cloud (OpenAI, Claude, Google, Mistral, OpenRouter, DeepSeek and more) |
The trade-off is complexity. SillyTavern has a lot of panels and settings, which is why this guide focuses on the few that matter.
Requirements
| Item | What you need |
|---|---|
| OS | Windows 10/11, macOS, Linux, Android (via Termux). Windows 7 cannot run Node.js 20. |
| Node.js | Version 20 or newer; the docs recommend the latest LTS |
| Git | For installing and updating |
| A model | Local back end or a cloud API key |
| For local models | A GPU with 8 GB+ VRAM is comfortable for 7-12B models; models take 5-50 GB of disk each |
SillyTavern itself is light. The hardware you need depends entirely on the model you connect.
Install on Windows
- Install Node.js LTS from nodejs.org and Git for Windows from git-scm.com.
- Create a folder Windows does not manage, for example
C:\AI\. Avoid Program Files, System32 and synced folders. - Click the File Explorer address bar, type
cmdand press Enter. - Run
git clone https://github.com/SillyTavern/SillyTavern -b release - Open the new
SillyTavernfolder and double-clickStart.bat. Do not run it as administrator. - The first run installs dependencies, then opens SillyTavern in your browser at
http://127.0.0.1:8000.
Prefer a menu-driven installer? The SillyTavern Launcher can install ST plus optional extras. Install Git with winget install -e --id Git.Git, then run:git clone https://github.com/SillyTavern/SillyTavern-Launcher.git && cd SillyTavern-Launcher && start installer.bat
The release branch is stable. The staging branch gets new features first but can break, so stick with release unless you want to test.
Install on Mac, Linux, Android and Docker
- Mac and Linux: install Node.js 20+ and Git, then
git clone https://github.com/SillyTavern/SillyTavern -b release,cd SillyTavern, and./start.sh. - Android: install Termux, then inside it install Git and Node.js and use the same clone and
bash start.shcommands. Pair it with a cloud API or a model running on your PC, as phones are slow for big models. - Docker: the repo has a
dockerfolder with a Compose file; rundocker compose up -dfrom there. The image isghcr.io/sillytavern/sillytavern.
Connect a Model
Click the plug icon (API Connections) at the top. You will choose between two API types:
| API type | How it works | Use with |
|---|---|---|
| Chat Completion | Sends messages as System / User / Assistant | OpenAI, Claude, Google, OpenRouter, DeepSeek and other cloud APIs; also local OpenAI-compatible servers |
| Text Completion | Sends one long text for the model to continue; needs the right instruct template | KoboldCpp, llama.cpp, Ollama, text-generation-webui, TabbyAPI |
| Back end | Default address | Notes |
|---|---|---|
| Ollama | http://127.0.0.1:11434 |
Easiest local option per the docs |
| KoboldCpp | http://127.0.0.1:5001 |
Single file, great for GGUF models |
| llama.cpp server | http://127.0.0.1:8080 |
For people who already use llama-server |
| text-generation-webui | http://127.0.0.1:5000 |
Start it with --api |
| AI Horde | None | Free community GPUs, no setup, slower |
The SillyTavern team notes they do not provide support for installing local back ends, so follow each back end’s own guide first and make sure it works on its own.
Step by Step: SillyTavern + Ollama
- Install Ollama and pull a model, for example
ollama pull llama3.1:8bor any chat model that fits your VRAM. Our Ollama guide covers this. - Make sure Ollama is running (the tray icon on Windows, or
ollama serve). - In SillyTavern, open API Connections and pick Text Completion.
- Set API Type to Ollama and the server URL to
http://127.0.0.1:11434. - Click Connect, then choose your model from the dropdown.
- Open Advanced Formatting (the “A” icon) and select the instruct and context template that matches your model family (for example Llama 3, ChatML or Mistral). Turning on auto-select helps.
Step by Step: SillyTavern + KoboldCpp
- Download KoboldCpp and a GGUF model, launch it, and wait for “Please connect to custom endpoint at http://localhost:5001”. Our KoboldCpp guide walks through it.
- In SillyTavern, choose Text Completion, API Type KoboldCpp.
- Enter
http://127.0.0.1:5001and click Connect. The model name appears when it works. - Set the matching instruct template in Advanced Formatting.
Not sure which GGUF file to download? Our GGUF quantization guide explains Q4_K_M, Q5_K_M and the rest.
Using a cloud API instead
Pick Chat Completion, choose the provider (OpenAI, Claude, Google AI Studio, OpenRouter, DeepSeek and so on), paste your API key and click Connect. Cloud models give the best quality with no GPU, but you pay per token and your chats go to that provider.
Create Your First Character
SillyTavern starts with a default assistant character. To make your own:
- Click the Character Management icon (the ID card at the top right), then Create New Character.
- Give it a name and an avatar image.
- Write the Description: who they are, how they speak and what they know. Keep it focused; 200-400 words is plenty.
- Write a First Message. This sets the tone more than anything else.
- Optionally add Example Dialogue under Advanced Definitions to show the style you want.
- Click Create and start chatting.
Ideas that work well: a Spanish tutor who corrects your mistakes gently, a fantasy game master who runs adventures, a job interviewer for practice, or a detective for a mystery story.
Characters are saved as PNG “character cards” with the data embedded, so you can import and share them. Only import cards from sources you trust and read them before use, because a card’s text goes straight into your prompt.
Set your own name and short description in Persona Management (the smiley icon) so characters know who they are talking to.
Key Settings to Get Good Replies
| Setting | Where | Tip |
|---|---|---|
| Context size | AI Response Configuration (sliders icon) | Match the size your back end loaded (for example 8192 or 16384); higher than the back end gives errors |
| Response length | Same panel | 250-400 tokens suits most chats |
| Temperature | Same panel | 0.7-1.0 for creative writing, lower for factual tasks |
| Min P | Same panel (Text Completion) | Around 0.05-0.1 is a good modern default |
| Instruct template | Advanced Formatting | Must match the model family or replies go strange |
| System prompt | Advanced Formatting | Short and clear instructions work best |
SillyTavern also ships with sampler presets. Start with a default preset for your back end and change one thing at a time.
Features Worth Trying
- World Info / lorebooks: store facts about places, people or rules. Each entry has trigger keywords and only enters the prompt when they appear, so long stories stay consistent without filling the context.
- Group chats: put several characters in one scene and let them take turns.
- Image generation: connect ComfyUI, Stable Diffusion WebUI or a cloud service to create a picture of the current scene. Our ComfyUI guide helps with the local setup.
- Text-to-speech: give characters voices using local or cloud TTS.
- Summarise: auto-summarise long chats so the model remembers earlier events.
- Data Bank: attach documents for retrieval, similar to “chat with your files”.
- Branches and swipes: swipe right to regenerate a reply, or branch a chat to try a different direction.
Update and Back Up
- Back up first: copy the
datafolder andconfig.yamlfrom your SillyTavern folder. Your characters, chats, lorebooks and settings live indata/default-user. - Windows: double-click
UpdateAndStart.bat. If it fails, open a terminal in the folder and rungit pull. - Mac/Linux/Termux:
cd SillyTavern,git pull, then./start.sh. - Docker: in the docker folder run
docker compose down, remove the old image, thendocker compose up -d.
Moving to a new install? Copy the data folder and config.yaml across. Do not copy the old public folder.
Troubleshooting
| Problem | Likely cause | Fix |
|---|---|---|
| Start.bat closes instantly | Node.js missing or too old | Install Node.js 20+ LTS, reopen the terminal, run node -v |
| Permission or EPERM errors | Installed in a Windows-controlled folder or run as admin | Move to a folder like C:\AI\ and run normally |
| “API connection failed” | Back end not running or wrong URL/port | Open the back end URL in your browser to check it is up |
| Replies are gibberish or talk for you | Wrong instruct template | Pick the template that matches the model family |
| Replies cut off | Response length too low | Raise response length or use Continue |
| Context errors | Context size larger than the back end loaded | Set the same context size in both |
| Very slow replies | Model too big for VRAM | Use a smaller model or lower quant |
| Update failed with git errors | Local file changes | Back up data, then follow the docs’ reset steps |
| Can’t open from another device | Listening only on localhost | Enable listening and whitelist in config.yaml, and keep it off the public internet |
My Take
If you already run local models and find basic chat apps limiting, SillyTavern is worth the learning curve. Characters, lorebooks and full prompt control make it the best tool I know for long creative writing sessions and language practice.
Start simple: Ollama or KoboldCpp, one 8-12B model, the default preset and one character. Once that feels good, add lorebooks, summarise and voices.
FAQ
Is SillyTavern free?
Yes. It is open source under AGPL-3.0. You only pay if you connect a paid cloud API.
Does SillyTavern include an AI model?
No. It is a front end. You connect it to a local back end like Ollama or KoboldCpp, or to a cloud API.
What is the easiest way to run SillyTavern locally?
Install Ollama, pull a model, then connect SillyTavern using Text Completion with the Ollama API type.
What port does SillyTavern use?
By default it opens at http://127.0.0.1:8000. You can change the port in config.yaml.
Can I use SillyTavern on my phone?
Yes, through Termux on Android, or by opening a SillyTavern server on your PC from your phone on the same network after enabling listening and the whitelist.
Can I use Claude or GPT in SillyTavern?
Yes. Choose Chat Completion, select the provider and paste your API key.
Where are my SillyTavern chats saved?
In the data/default-user folder inside your SillyTavern install. Back up that folder and config.yaml.
Release or staging branch?
Use release for stability. Staging gets new features first but can have bugs.




