If you’ve seen “SillyTavern” mentioned in AI communities and can’t tell if it’s an app, a website, or something you have to build yourself, that confusion is common, because it’s technically none of the above. SillyTavern AI is a free, open-source interface that lets you chat with AI characters using your choice of AI model in the background. It doesn’t come with its own AI built in, and it’s not something you download from an app store and open, it’s a small local server you install and run yourself, then connect to a model of your choosing.
This guide is for actual beginners, no prior coding background assumed, covering exactly how to use SillyTavern AI from a completely blank starting point, including how to run SillyTavern AI Android setups specifically. It walks through what SillyTavern actually is, how to install it on Windows, Mac, or Linux, how to connect an AI model, how to get it running on Android, how to access it remotely, how character cards work, and what to do when something breaks. Every step here is checked against the official SillyTavern documentation rather than guessed at, since installation steps and requirements change with new releases.
What Is SillyTavern and What Is It Used For?
SillyTavern is an open-source AI chat frontend, a self-hosted chatbot interface you install on your own computer or phone, which then connects to an AI model of your choice for character-based chat and storytelling. It is not itself an AI model. Think of it as the AI roleplay interface and the memory system, the character cards, chat history, personas, and formatting, while the actual “brain” generating responses comes from whatever AI backend you plug into it separately.
People use it for a wide range of things: collaborative fiction and storytelling, D&D-style tabletop roleplay with an AI game master, chatting with a custom-built assistant persona, or just experimenting with how different AI models respond to the same character setup. It started as a fork of an older project called TavernAI and has been actively developed since, now maintained by a large volunteer community on GitHub with regular releases. It’s worth noting up front that this guide sticks to the general-audience side of that range, installation, setup, and character basics, the same way a guide to a general-purpose tool like a video editor or a drawing app would, without getting into any specific content genre.
It’s also worth placing SillyTavern next to a different kind of AI product people have been hearing about lately: a hosted, action-taking assistant like Meta’s Muse AI agent. Both let you interact with an AI persona of sorts, but the structure is nearly opposite. Muse runs entirely on Meta’s servers and can take real-world actions like booking or shopping on your behalf. SillyTavern runs on your own machine, does nothing beyond generating conversation text, and requires you to bring your own model. If Meta Muse is a hosted agent that acts for you, SillyTavern is closer to a workbench you assemble yourself for a much narrower job, chatting with a character you control.
The core appeal, and also the source of most of the setup friction people run into, is that SillyTavern gives you far more control than a hosted sillytavern ai chat app would through a typical closed platform. As a large language model frontend, it lets you choose the AI model, control how much conversation history it remembers, write a detailed prompt template and character personality, and set up a persona chat where you define your own role in the story as well as the character’s.
Your data stays on your own machine rather than living on someone else’s server. The tradeoff is that “choose your own model” means an extra setup step a hosted app wouldn’t require, which is exactly what the rest of this guide walks through.
Is SillyTavern AI Free?
Yes, the SillyTavern software itself is completely free and open-source, with no subscription, license fee, or paid tier of any kind. You can download it, install it, and use it without ever paying SillyTavern’s developers anything.
What isn’t automatically free is the AI model you connect it to, since SillyTavern doesn’t generate responses on its own. If you connect a paid API, like a commercial cloud model billed per use, you’ll pay whatever that provider charges. If you connect a model with a genuinely free tier, or run an open-source model locally on your own hardware, using SillyTavern can cost you nothing beyond the free software itself. The cost question really has two separate answers: the interface is free, full stop, and the AI behind it may or may not be, depending entirely on what you choose to connect.
How to Download and Install SillyTavern
Requirements
You need two things before you start: Node.js, specifically version 20 or newer, since current SillyTavern releases expect a modern Node environment and older versions can cause confusing install failures, and Git, which is used to download the official project files and pull future updates with a single command rather than manually re-downloading everything.
You’ll also want a normal, user-owned folder to install into. Avoid installing inside Program Files, System32, or any other permission-locked system directory on Windows specifically, since Windows’ file permissions there commonly cause install and update failures that look like bugs in SillyTavern but are actually a permissions issue with the folder location.
Step-by-Step Install (Windows, Mac, Linux)
1. Install Node.js. Go to the official Node.js website and download the current LTS (long-term support) version, then run the installer accepting the default options.
2. Install Git. Download it from the official Git website and install it with the default settings.
3. Verify both installed correctly. Open a terminal (Command Prompt on Windows, Terminal on Mac or Linux) and type node --version, then git --version. Both should return a version number. If either doesn’t, revisit that installation step before continuing.
4. Download SillyTavern. Navigate to the folder where you want to install it, then run:
git clone https://github.com/SillyTavern/SillyTavern -b release
This pulls the release branch, which is the stable, recommended version for most users. A separate staging branch exists with newer features, but it can break unexpectedly and is really meant for more advanced users comfortable troubleshooting on their own. If you’d rather skip Git entirely, you can download the latest release as a zip file directly from the project’s GitHub releases page instead, though you’ll then be manually re-downloading and merging your settings for every future update rather than running a single update command.
5. Start the server. On Windows, open the SillyTavern folder you just downloaded and double-click Start.bat. On Mac or Linux, open a terminal in that folder and run ./start.sh. The first run automatically installs all the required dependencies, which can take a few minutes depending on your connection.
6. Open it in your browser. Once the server finishes starting, it will either open automatically or tell you to visit localhost:8000 in your browser. That’s it, the interface loads locally and you’re looking at a running SillyTavern instance with no AI model connected yet.
A few common first-run snags worth knowing about ahead of time: an old, outdated Node.js version is a frequent, silent cause of install failures, so if you installed Node a while back for something else, update it before troubleshooting anything else. A flaky internet connection during the automatic first-run dependency install can also cause it to fail partway through, in which case simply closing the window and re-running the start script picks up where it left off rather than starting over.
How to Connect an AI Model to SillyTavern
This is the step that trips up most beginners, not because it’s complicated, but because it’s the one piece SillyTavern deliberately doesn’t handle for you. SillyTavern runs on a “bring your own backend” model: you pick an AI provider, generate an API key or run a local model server, and plug that connection into SillyTavern’s settings.
Inside SillyTavern, click the plug icon in the top navigation bar to open the API Connections panel. From there, you’ll choose an API type, this could be a commercial cloud API like Google AI Studio or Google’s Vertex AI, or a local AI hosting setup running on your own computer, and enter whatever credentials or connection details that option requires, most commonly an API key you generate from the provider’s own website. Once connected and tested, SillyTavern will use that model to generate every response, while it continues to handle the character card, chat memory, and formatting on its own side.
If you’d rather not use a cloud API at all, local backends like KoboldAI let you run an open-source language model directly on your own hardware, with SillyTavern connecting to it over your local network instead of the internet. This avoids ongoing usage costs entirely but requires hardware capable of running the model you choose, and generally runs slower than a well-resourced cloud API.
What’s the Best AI Model for SillyTavern?
There’s no single universally “best” model here, it depends on what you’re optimizing for, and it’s worth thinking about the trade-offs rather than chasing a specific ranked recommendation.
Cost varies enormously between providers and between free and paid tiers of the same provider. Some options offer a genuinely usable free tier for light, occasional use, while heavier or more frequent use tends to push you toward a paid plan billed by usage.
Context length, meaning how much of your ongoing conversation and character information the model can actually keep in mind at once, matters a lot for longer roleplay or story sessions specifically, since a model with a short context window will start “forgetting” earlier details in a long-running chat sooner than one with a longer window.
Speed differs based on both the model itself and, for local backends, your own hardware. A cloud API generally responds faster and more consistently than a local model running on a modest computer, though a local setup gives you more privacy and no per-use cost.
The practical approach for a beginner is to start with whichever option has the lowest barrier to entry, generally a cloud API with a free tier, get a character chatting successfully, and only optimize further once you have a concrete sense of what you actually want more of: speed, memory length, or lower cost. If you’re comparing providers by searching around for documentation and pricing pages, and Google’s AI-generated summary box keeps getting in the way of the actual source pages you’re trying to read, our guide on removing Google’s AI Overview covers a couple of quick fixes for that.
How to Use SillyTavern on Android or Mobile
There is no official standalone SillyTavern mobile app, and no official APK. This is worth stating plainly, because searches for a “SillyTavern APK” turn up plenty of unofficial third-party files claiming to be one, and downloading an app package from outside the Google Play Store or the official SillyTavern GitHub carries real security risk, since there’s no way to verify what’s actually bundled inside it. Stick to the official installation routes below.
The reason there’s no simple app in the first place comes down to what SillyTavern actually is under the hood: a Node.js server, not a self-contained mobile app. That server needs somewhere to actually run, and Android’s aggressive battery management treats long-running background processes, exactly what a local server is, as something to shut down to save power. That single fact shapes every real option for using SillyTavern on a phone.
Running it directly on your Android device with Termux is the official, community-documented route. Termux provides a Linux-style command line environment on Android capable of running the same Node.js server SillyTavern uses on desktop. Important: install Termux from F-Droid or directly from its GitHub releases page, not the Google Play Store version, which is outdated and no longer maintained. Once Termux is installed, the setup is a short handful of commands:
pkg update && pkg upgrade
pkg install git nodejs
git clone https://github.com/SillyTavern/SillyTavern -b release
cd SillyTavern && ./start.sh
Then open localhost:8000 in your phone’s browser, same as the desktop version. If you’re on a 32-bit Android device specifically, you’ll need one extra dependency (pkg install esbuild) before the standard steps will complete successfully, since some packages can’t be installed through npm alone on that architecture.
A couple of practical notes for running it this way: Android can still put Termux to sleep in the background even after it’s running, which will interrupt your server, so using Termux’s wake-lock feature to prevent deep sleep helps keep a session running while you’re actively using it. It’s also worth trimming SillyTavern’s resource use on a phone specifically, adjusting a few settings in its config.yaml file, like disabling automatic chat backups and reducing character data caching, can meaningfully help performance and storage use on a device with far less power than a desktop computer.
Accessing a desktop-hosted instance from your phone’s browser is the other mainstream option, and it’s simpler if you already have SillyTavern running on a PC. Rather than installing anything on the phone at all, you connect to your computer’s SillyTavern server from your phone’s browser over the same WiFi network, or remotely using the methods in the next section. This shifts all the actual processing to your computer and leaves your phone as just a browser window into it.
You may also come across community-built Android launcher apps that wrap the entire Termux-plus-SillyTavern setup into a single installable app with a friendlier interface. These aren’t part of the official SillyTavern project, they’re independent, community-maintained tools, so treat them the same way you would any third-party app: check that it comes from a visible, actively maintained open-source repository before installing anything, rather than trusting an APK from an unfamiliar site.
How to Access SillyTavern Remotely
By default, SillyTavern only accepts connections from the same machine it’s running on, which is why accessing it from a phone or another device takes an extra step even on your own home network.
For access within your own home network (using SillyTavern on your phone while it runs on a PC in the same house), you need to set listen: true in SillyTavern’s config.yaml configuration file, which tells the server to accept connections from other devices on the same network rather than only itself. Once that’s set and the server is restarted, you can reach it from another device on the same WiFi by entering your computer’s local network IP address followed by the port, instead of localhost.
For access from outside your home network entirely, SillyTavern’s own official documentation is explicit and worth repeating directly: don’t use router port forwarding to expose your instance to the open internet. Instead, use a VPN service like Tailscale, or a tunneling service like Cloudflare Tunnel or ngrok. SillyTavern even ships a built-in script (Remote-Link.cmd on Windows) that automatically sets up a temporary Cloudflare tunnel for you, printing out a temporary web address you can use to reach your instance from anywhere while the tunnel stays active.
Whichever remote method you use, treat this step with real caution. SillyTavern’s own documentation carries an explicit warning against hosting an instance on the open internet without proper security in place, since your instance holds your chat history, character data, and any API keys you’ve connected. At minimum, set a username and password in config.yaml before opening any kind of remote or tunneled access, and treat any tunnel URL you generate as something to keep private rather than share.
Character Cards: Where They’re Stored and How to Add One
A character card is what actually gives an AI persona its personality, background, and conversational style inside SillyTavern, bundling a description, a scenario, sample dialogue, and other structured details into a single file the software can read.
Where Character Cards Are Stored
Character cards live inside your SillyTavern installation’s data folder, specifically under a path like data/default-user/characters on most setups. Cards themselves are most commonly PNG image files with the character’s actual data embedded invisibly inside the image file, alongside plain JSON files in the newer card formats. This is why a character card can double as its own preview picture and a portable data file at the same time.
How to Add or Create a Character Card
To add an existing card, open the Characters panel in SillyTavern (usually represented by a character icon in the left-hand navigation) and use the import option to select a card file from your computer. If it’s a properly formatted PNG or JSON card, it should load in with the character’s name, description, and avatar already filled in, ready for a new chat.
To create your own from scratch, the same Characters panel has an option to build a new character manually, where you fill in fields for the character’s name, personality description, an opening scenario, and example dialogue showing how they talk. You don’t need every field filled in perfectly on your first attempt, a short description and one solid example message is usually enough to get a workable starting point you can refine as you chat.
If you want a custom avatar image to go with a character rather than a placeholder, SillyTavern supports connecting an image generation backend like Stable Diffusion through its extensions panel, and the same general prompt-writing principles covered in our guides on the 1980s AI photo trend and AI Prompt Ghaus Editz styles, describing lighting, styling, and composition specifically rather than vaguely, carry over directly to generating a consistent character portrait here too.
How to Import Janitor AI Character Cards into SillyTavern
Janitor AI and SillyTavern both use a compatible card format at their core (commonly called the V2 or V3 Tavern card spec), which means a character built on one can generally be brought into the other, though the export step is where most people run into friction rather than the format itself.
The first step is getting the character out of Janitor AI as an actual file rather than just viewing it in your browser, look for a download, export, or save option on the character’s page. What you’re aiming for is either a PNG card with the character data embedded in it, or a JSON file containing the same information. It’s worth knowing that Janitor AI’s site runs behind bot-protection infrastructure that can occasionally interfere with a clean PNG export or import, a known, documented issue in SillyTavern’s own GitHub tracker, so if a direct PNG doesn’t import cleanly on the first try, don’t assume you’ve done something wrong.
A dependable fallback in that situation is copying the character’s description, scenario, and example dialogue fields manually into a new character you build directly inside SillyTavern, which sidesteps the file-format issue entirely at the cost of a few extra minutes of copy-pasting.
If you do get a working file, importing it into SillyTavern uses the exact same Characters panel import option covered above, there’s no separate “Janitor AI import” button, it’s the same general character import process regardless of which platform the card originally came from.
SillyTavern vs. Janitor AI vs. Character AI: What’s the Difference?
These three come up together constantly, and the confusion is understandable since all three let you chat with an AI character. The actual difference is structural, not a matter of one being simply “better” than the others.
SillyTavern is a self-hosted frontend. You install it yourself, you provide your own AI model or API connection, and your character data and chat history live on your own device by default. Nothing about it works without you separately connecting a backend model first.
Janitor AI and Character AI are both hosted platforms, meaning you sign up for an account, and the AI model runs on their servers rather than yours, with no separate model connection or technical setup required on your end. You get a simpler, faster on-ramp at the cost of less control over exactly which model is generating responses, how much conversation history it retains, and where your data is stored.
Neither structure is objectively superior, they’re built for different priorities. A hosted platform trades control for convenience, letting you start chatting within minutes of signing up. A self-hosted frontend like SillyTavern trades that convenience for control, more customization, choice of model, and data staying local, at the cost of the setup process this entire guide has walked through. Which one makes sense depends entirely on whether you value getting started immediately more, or having full control over the model and your data more. Whichever route you choose, each platform sets its own content and usage rules, and it’s worth reading a platform’s terms of service directly rather than assuming one platform’s policies apply to another.
Can I Use SillyTavern with Discord?
There’s no official Discord integration built into SillyTavern itself, and it’s worth knowing that directly rather than assuming a missing feature is a bug. SillyTavern’s own developers have stated plainly that the project is intended to function purely as a frontend, without a built-in external API for other apps like a Discord bot to hook into directly.
What does exist are independent, community-built bridge extensions that connect a separate Discord bot to a running SillyTavern instance, letting messages flow between a Discord server and your character chat. These are third-party projects maintained by individual community developers, not the SillyTavern core team, so treat them the way you would any third-party extension: check that the repository is actively maintained and comes from a visible GitHub source before installing it, and don’t expect the same level of long-term support or documentation the core project itself has.
Common Problems: SillyTavern Not Working, Login Issues, and Fixes
The server won’t start, or dependencies fail to install. This is usually either an outdated Node.js version or a flaky connection during the automatic first-run install. Confirm node --version shows 20 or newer, and if it stalls partway through installing dependencies, close the window and re-run the start script, since it resumes from where it left off rather than starting completely over.
Nothing happens when a character sends a message, or you get an error instead of a reply. This is almost never actually a SillyTavern bug, it’s an API connection problem. Double-check your API key is entered correctly and hasn’t expired, confirm the base URL and model name match exactly what your provider expects, and if you’re using a local backend, confirm that backend server is actually running and reachable before assuming SillyTavern itself is broken. The most common single cause of this specific problem is an expired or malformed API key, not the software.
Install or update fails specifically on Windows. Check whether SillyTavern is installed inside a permission-locked folder like Program Files or System32. Windows’ file permissions in those locations cause exactly this kind of failure, and it looks identical to a genuine software bug until you move the installation to a normal user-owned folder and try again.
A character card won’t import, or imports with garbled text. A card built for a newer card specification (V3) can sometimes fail on an older SillyTavern version expecting the older V2 format, updating SillyTavern to the current release usually resolves this. Garbled or mojibake-style text in a character’s dialogue is typically an encoding issue introduced somewhere in how the card was originally exported or re-shared, re-exporting or re-saving the card through a tool that writes clean UTF-8 text generally fixes it.
If a PNG card was re-shared through a platform that recompresses images, including some Discord attachments, the embedded character data itself can be stripped out entirely by the recompression, so always try to get a card from its original source rather than a screenshot or a re-upload.
You can’t reach SillyTavern from your phone even on the same WiFi. Confirm listen: true is actually set in config.yaml and that you restarted the server after changing it, and make sure you’re using your computer’s local network IP address rather than localhost, since localhost only ever refers to the device you type it into.
Frequently Asked Questions
Is SillyTavern safe to install?
The core SillyTavern project itself is open-source, actively maintained, and hosted on GitHub where anyone can review the code, which is a meaningfully different trust situation from an unverified app. The real safety consideration isn’t the core software, it’s what you add to it: only install third-party extensions and character cards from sources you can verify, and only download Termux or SillyTavern itself from the official F-Droid, GitHub, or project website links covered in this guide rather than a random third-party mirror.
Does SillyTavern work on iPhone?
There’s no official iPhone or iPad app, and the Termux route covered in this guide is Android-specific, since Termux itself doesn’t have an iOS equivalent due to Apple’s more restrictive app sandboxing. The realistic option on iPhone is accessing a SillyTavern instance already running on a separate computer, from Safari or Chrome on your phone, using the local network or remote access methods covered earlier in this guide.
Does SillyTavern store my chats and data online?
By default, no. Your character data, chat history, and settings are stored locally in files on whichever device is actually running the SillyTavern server, not on a cloud service SillyTavern’s developers control. The one exception is whatever AI model you’ve connected, since your messages do get sent to that provider’s servers to generate a response, governed by that specific provider’s own data and privacy policies rather than SillyTavern’s.
Can more than one person use the same SillyTavern install?
Yes, SillyTavern supports multiple user profiles on a single installation, each with its own separate characters, chat history, and settings, which is useful if more than one person shares the same computer or server instance. This is set up through the User Profiles section in settings, and it’s separate from the network access settings covered earlier, which control whether other devices can reach the server at all.
Do I need to know how to code to use SillyTavern?
No, the day-to-day experience of chatting, creating a character card, and adjusting settings is entirely point-and-click through the browser interface once it’s running. The parts that can feel technical, and where most of this guide’s actual instructions are focused, are the one-time setup: installing Node.js and Git, running a couple of terminal commands, and pasting in an API key. None of it requires writing code yourself, just following the specific steps once.
Get Chatting
SillyTavern’s setup process has more steps than a hosted chat app, but each one is doing something specific: Node.js and Git get the software onto your machine, an API connection gives it something to actually generate responses, and a character card gives that AI a personality to respond as. Install it, connect one model, and try a starter character card before worrying about tunnels, Android setups, or Discord bridges, those are all genuinely optional extras once the core loop is working.
For more practical breakdowns like this on AI tools and how to actually get them running, check the blog or visit the homepage for more guides.
Installation steps, version requirements, and official documentation links referenced here reflect SillyTavern’s official documentation and GitHub repository as of publication. Since this is an actively developed open-source project, always check the official docs directly if a step here doesn’t match what you see on screen.
