Here is what people ask before and after installing. None of it needs a terminal, and if a word is new to you, the list at the bottom says what it means.
Yes. SillySimple is free: nothing to pay, no subscription, no account. The models it downloads to your computer are free too.
The one thing that can cost money is an online service, if you choose one to write the replies or to read them aloud: that service may charge you, on its side, for the key you give the app.
Yes. SillySimple is free software, under the GNU General Public License, version 3 (GPL-3.0). Anyone may read its code, check what it does, change it and share it; a changed version that is shared must stay open, under the same licence.
The name SillySimple and its icon are not part of that: a changed version goes by another name.
The code is not public yet. This page will link to it when it is.
To run a model on your own computer: a Mac with an Apple chip, or a Windows or Linux PC with a graphics card. Then 8 GB of memory, and room on the disk for a model: about 5 GB for the smallest one on a Mac, and a few more on Windows and Linux, where the engine that runs models is larger.
On another computer, an online service can write the replies, with a key of your own. Nothing is downloaded then, and the computer hardly matters.
No terminal, and nothing to install besides the app.
On a Mac with an Apple chip, no: the chip does that work. On a PC, yes, if the model is to run on your computer: a graphics card is what makes replies come in a few seconds.
On a PC without one, nothing is promised. The model may still run on the processor alone, but replies can be very slow, and this has not been tried. There, let an online service write the replies, with a key of your own: see “Can replies come from Claude, OpenAI, Mistral or OpenRouter?” below.
So far, SillySimple has run a model on one kind of computer only: a Mac with an Apple chip.
Yes, once a model is on your computer. Downloading that model is the only moment the app needs the internet, unless you choose an online service to write the replies.
The app looks at how much memory your computer has, and offers what fits among three sizes of Gemma 4. Small (E2B) is a 4.6 GB download and is offered to every computer. Medium (12B) is 8 GB, for computers with 16 GB of memory. Large (26B A4B) is 18 GB, for computers with 32 GB.
The largest one that fits is marked “Recommended”. A larger model writes better, more slowly.
Yes. SillySimple runs models with Ollama, and Ollama shows as several lines: a small one for the program itself, and one for each model held in memory. They are named “ollama” or “llama-server”.
The biggest is the model that writes the replies: the Gemma 4 you picked (Small, Medium or Large), or the model you chose yourself. On a Mac with an Apple chip there can be a second, much smaller one: EmbeddingGemma 2, by Google (270 million parameters, a 378 MB download), which reads the mood of each reply. On Windows, Linux and Intel Macs it is not used, so there are two lines at most.
When SillySimple runs its own Ollama, a model leaves memory half an hour after it was last used, and everything stops when you quit the app. Your own Ollama, if you have one, keeps its own settings.
The model is being loaded into memory, which takes a moment for a file of several gigabytes. The replies that follow come faster. The model stays in memory for half an hour after the last reply.
If Ollama is open on your computer, SillySimple uses it instead of starting a second one, and both keep their models in the same place: what one downloaded is there for the other.
For LM Studio, KoboldCpp, llama.cpp or any server that speaks like OpenAI’s, choose “Another server” and give its address.
Yes, with a key of your own from that service: choose “Online service”. Nothing is downloaded, and replies are quick on any computer.
The conversation and the character’s card are then sent to that service to write each reply, instead of staying on your computer.
Yes. Under the list, “Another model…” downloads any model of ollama.com by its name, or a GGUF file from Hugging Face.
No filter added by SillySimple: what a model accepts is up to the model you choose. With a model on your computer, everything stays on it.
An online service applies its own rules to what it is sent.
In Settings › Connection, the bin at the end of a model’s line removes it from the disk, after asking you to confirm. You can download it again later.
In one folder on your computer. Characters are card files in the format SillyTavern uses, chats are in a database, settings in a file. Settings › Data & Privacy says what is kept and opens the folder.
On a Mac it is ~/Library/Application Support/SillySimple; on Windows, %APPDATA%\SillySimple.
With a model on your computer: nothing you write. There is no account and no SillySimple server, so nothing is ever sent to us.
With an online service, the conversation and the character’s card are sent to that service to write each reply. If you turn on reading aloud with an online voice, the text of a reply is sent to that voice service when it is read.
Settings › Data & Privacy saves everything into one .zip file, which you can open by hand. On the other computer, restore it from the same page: what the file holds is shown before you confirm.
API keys are never part of a backup: they are encrypted by the computer they were typed on.
Settings › Data & Privacy has “Erase everything”, which puts the app back as it was on the first day. You can also delete all chats and keep your characters.
What your persona does goes between asterisks, what they say between quotes: *I sit down by the window.* "Tea, please." The app shows actions in italics, and models know the habit.
To say something to the AI itself, outside the story, write it in brackets after OOC (out of character): (OOC: write longer, more detailed replies). The OOC button in the message box writes the brackets for you. It works at once, and the line stays in the chat: delete it once it has worked if you like.
For something that should hold for the whole chat, the app has steadier ways: Reply length in the right panel for how long replies are, and the system prompt (Settings › Generation, or the character’s own) for standing instructions.
It ran into the length limit. Press Continue under the message to get the rest, or make replies longer with Reply length, in the right panel.
Open the Memory tab in the right panel: it shows what is kept of the older messages. Add what is missing, in a sentence. For things that must never be forgotten, write them in your Persona or in the character.
The app notes where the scene is after each reply, and tells the character to stay there until you go somewhere else. The note is at the top of the Memory tab: correct it if it is wrong, and the next reply follows it.
Edit the reply and cut that part, or write another version with the arrows. Models copy what they see: a first message that never speaks for you sets the tone.
With a model on your computer, speed depends on the computer and on the size of the model. Shorter replies come sooner; a smaller model is faster; an online service is usually the fastest.
The dot at the bottom left says whether the app reaches the model. If it is not green, open Settings › Connection: no model may be downloaded yet, or the server you chose (LM Studio…) may not be running.
Up to four. “Add a character”, at the top of a chat, brings another one in. They answer in turn, or you click a name to make that character speak: they can talk to each other.
Yes, as an option that is off until you turn it on, in Settings › Voice. It takes a key from ElevenLabs, or a voice server in OpenAI’s format. The text of a reply is sent to that service when it is read.
Yes. Drop their PNG or JSON cards into the Library: the three versions of the card format are read. Their greetings, their lorebook and the favourite mark come with them.
They do not come along. SillySimple has no extensions, and far fewer settings: the memory of long chats, the notes on the scene and the setup of the model are part of the app, already on.
No. It is written from scratch and is an independent project, not affiliated with SillyTavern. It reads the same character cards.
It owes a lot to SillyTavern, though. I made SillySimple after using SillyTavern myself, and I have a great deal of respect for the work of its team. SillySimple exists to make the same kind of chat easy to start, not to replace it.
