A whole tavern,
and its keeper.
PokiTavern runs the real SillyTavern on your Android phone. No Termux, no PC, no account: the inference engine is in the app, so once you download a free model it replies in airplane mode, with nothing sent to an AI vendor and no token bills. Want a bigger model? Plug in your own key.
A SillyTavern that thinks on its own.The official 1.18 frontend, a compatible server and on-device inference, in one Android app.
The real 1.18 frontend, served pixel-for-pixel from an embedded server that speaks the exact SillyTavern API. Add a model on-device and it writes replies with the network off. Front end, server and model all live on your phone.
Its own brain, no cloud
Pick a model from the list inside the app: Qwen3 0.6B at about 614 MB, or Gemma 4 E2B at about 2.6 GB. Download it once and it runs on the phone itself, with no token bills and no network.
The real SillyTavern
Not a clone. The app serves the official SillyTavern 1.18 frontend, so your cards, presets, world books and extensions behave as they do on desktop, because the embedded server matches the original API field by field.
Want it smarter? Bring a key
On-device is free, private and offline. When you want more, point it at OpenAI-, Claude-, Gemini- or OpenRouter-compatible endpoints. They are the same connections you already use on desktop.
Private by default
Characters, chats, presets and world books live in a local SQLite store on your phone. With the on-device model, prompts are never sent to any AI vendor. Nothing leaves unless you say so.
Serve your whole LAN
Your phone becomes the tavern. Open its Wi-Fi address on a laptop or tablet and keep every chat in one place.
Painless updates
The app checks for new releases itself, downloads over a CDN and verifies every byte with sha256 before installing.
The bar's open, signal or not.Offline AI roleplay that answers in airplane mode.
Every other way to get SillyTavern on a phone needs a PC left running at home, a tunnel to it, or a cloud GPU by the hour. PokiTavern runs the model on the phone itself: no token bills, no accounts, and nothing sent to an AI vendor.
It's a light pour, best for a quick chat on a plane or the subway. When you want a heavier one, add your API key and the same tavern routes to a full-size model.
One phone, every screen in the house.Serve SillyTavern over your LAN from an Android phone.
PokiTavern listens on your Wi-Fi network, not just on the phone itself. Tap the floating ball, copy the LAN address, something like http://192.168.1.23:41520/, and open it in any browser on the same network.
Same characters, same chats, same settings everywhere, because there is only one server: the one in your pocket.
Pouring in three stepsHow to install SillyTavern on Android without Termux.
Install the APK
Grab the release below and install it. Android will ask you to allow installs from your browser, which is normal for any app distributed outside the Play Store.
Pick your pour
Download a model for free offline chats, or paste an API key for a full-size one. Do both and switch anytime; it is the same tavern either way.
Import and chat
Import your cards, start chatting, and open the LAN address on a bigger screen if you like.
Before you pull the tapCommon questions about running SillyTavern on a phone.
How is PokiTavern related to the SillyTavern project?
It is not affiliated with it. PokiTavern embeds a server that implements the SillyTavern 1.18 API and serves the project's official frontend on top. SillyTavern is a separate open-source project and has not endorsed this app.
Is this the real SillyTavern frontend or a lookalike?
The real one. The app serves SillyTavern's official 1.18 web frontend from a server running inside the phone, so presets, themes and extensions behave as they do on desktop. The API underneath matches the original field by field.
Do I need Termux or a PC?
Neither. A Termux install needs a terminal, Node.js and a git clone, and Android can reclaim the session and stop the server. PokiTavern is an ordinary Android app: install it, open it, and the server is already listening.
Does it really work offline, and what model does it use?
Yes, once a model is on the phone. Pick one from the list inside the app, Qwen3 0.6B at about 614 MB or Gemma 4 E2B at about 2.6 GB, download it over Wi-Fi once, and it runs through LiteRT on the device. After that airplane mode works.
Can I import character cards and world books?
Yes. PNG cards carry their JSON in the image's tEXt chunk: V2 under the chara keyword, V3 under ccv3, with V3 taking precedence when both are present. PokiTavern reads both, as does SillyTavern 1.18. World books import the same way.
Do I need an API key?
No. The on-device model needs no key and no account. A key is optional: point the app at OpenAI-, Claude-, Gemini- or OpenRouter-compatible endpoints when you want a larger model than the phone can hold.
Is it free? Is there a message limit?
Free, with no message quota, no subscription and no account. If you connect your own API key you pay that provider directly, at their prices. PokiTavern never sits between you and the model you chose.
Why is the download around 200 MB?
Because the app carries the on-device inference runtime, not just a user interface. Version 1.1.2 is 202 MB: the LiteRT engine and its GPU and NPU backends take most of it, and the bundled SillyTavern frontend adds about 15 MB. Models download separately.
When is the iOS version coming?
It is in progress and not yet on the App Store. The waitlist below is the only notice we send, and we use the address for nothing else.
Does anything I type get uploaded?
Not with the on-device model: prompts never leave the phone. Attach your own API key and prompts go to that provider, and nowhere else. There is no PokiTavern account and no telemetry server to receive them.
Pull up a chairDownload PokiTavern for Android.
Requires Android 7.0+. Updates arrive in-app, so you only ever do this once.
The iOS app is coming to the App Store. Leave your email and we'll send you an invite the moment it opens.
One email, one invite. See our privacy policy.