z's super basic
SillyTavern Guide
for assistant-as-companion builds

This guide is for setting up your SillyTavern similarly to a CustomGPT. This is not like a classic RP SillyTavern build with swappable character cards.

This build is for anyone who likes the relationship-with-the-AI-assistant-directly approach, which is probably most ChatGPT users. This build works for single or multiple companions.

If you have multiple companions that do not normally interact with each other, this guide might not be optimal, and a more standard SillyTavern build would probably suit your use case.


Multiple Companion note: for multiple companions that normally interact with each other, build the main companion first and consider lorebook entries to trigger the others. 

How SillyTavern Works

SillyTavern is a free and open source chat interface that specializes in managing context for roleplaying.

You only really need your API working and parameters set for the LLM to start talking. Everything else is just configuration and quality of life upgrades.

Connecting your API

Before messing with settings, follow the five steps in the official SillyTavern guide for connecting your API: 
https://docs.sillytavern.app/usage/api-connections/openrouter/

If you do not connect your API none of this will work.

Main Settings

First, we are going into the AI Response Configuration tab.

We are going to pay attention to the following settings:

Context Size
Set it to 64000.
Even if your model has a huge context window capability, this is often the tried-and-true context size for complex interactions. This size does well with keeping things coherent, without mixing things up. Feel free to mess around with this later.

Max Response Length
Set it to 2500.
You can set it higher, but you want some kind of limit, in case the model glitches and starts spewing endless amounts of text. Setting it lower might also end up with you having a long and thoughtful response cut off.

Streaming
Check this box.
Streaming is when the words appear as they're generated. Not all APIs support this, and will only display the full message when it is completely generated. This is optional, but most people like streaming.

Parameters
Set them according to the recomendations for the model.
You can google "reddit SillyTavern <full model name here> recommended parameters" or go to the model's Huggingface page.
The model's full name is important. Meaning, if using GLM 4.7, google "reddit SillyTavern GLM 4.7 recommended parameters" not just "GLM parameters" because they vary between the individual model. SillyTavern users will have parameter settings more relevant to our use case, as sometimes the parameters used for coding are different.
If there are more parameter sliders in the SillyTavern settings than you find recommendations for, then just leave the extra ones alone :) Just don't touch em, they're fine.

Save your settings!

Next, scroll down to the bottom of the settings.

We are going to be using the Main Prompt to store your companion's custom instructions.

If relevant, put your NSFW jailbreak here as well. Different models need different jailbreaks, and there are different techniques.

Save!
Click the little save-button. This one is only for saving the Main Prompt, clicking this will not save all of your settings.

After you've filled out your Main Prompt, scroll back up to the top of your settings and save those again too!

Persona Management

Now we navigate to the persona management tab.
This is where information about the user (you) is stored.

Create a new persona. Make sure it's selected, and then put all the information you want your companion to know about you in the third-person.

Example: Z is a woman with dark brown hair. She loves symbolic and surrealist artwork. When she is sad she likes really big hugs.

The persona should be set as the default.

Character Management

Next, we go to the character management tab.

Note: in traditional RP setups, this is where character prompts would normally go. But, for this use case, we'll use it for something different.

Create a new character, then select it.

Once we're in the character details tab, write your companion's visual description here.
The way SillyTavern is set up, this is a helpful place to store visual details.

Select the "More..." dropdown box.

Set your companion as the Welcome Page Assistant. 

Extensions

Now we'll navigate to the Extensions tab.

Summarizer

The summarizer is very helpful for keeping a chat coherent when it gets long. Click on the Summarize dropdown menu.

Each summary that is generated in a chat is tied to that chat, you don't need to worry about anything getting messed up when you switch chats.

In the "Summarize with" dropdown box, make sure Main API is selected.

Next, click Summary Settings.

Set the Target summary length to 1000.

Close the summarize dropdown.

Vector Storage

Now open the Vector Storage dropdown menu.

Vector Storage deals with vectorization, which is how we make data RAG compatible. RAG is one of the main ways pieces of information are retrieved and put into context.

In the Vectorization Source dropdown box, make sure Local (Transformers) is selected.

Check Enable for files, this allows you to attach files in a data bank for your companion to read from.

Check Enabled for chat messages, this allows messages in a chat that are past the context window to be referenced as needed.

After this is done, click Vectorize All to test that it is working.

You should see this green pop-up to let you know vectorization is working. You don't really need to mess with this again, it should work automatically from here.

Data Bank

The data bank is where you can store files to be vectorized, much like knowledge files in CustomGPTs.

This would be where you might store external documents, like long term memory files.

To access it, click on the magic wand button at the bottom of the SillyTavern window in the message bar, then select Open Data Bank.

To add files that your companion can access across all chats, click the Add button in Global Attachments.

World Info/Lorebooks

Now we'll move on to World Info tab.

World Info, also known as lorebooks, are a way to insert prompts into context by using keyword triggers. This is very helpful for storing things that you don't need your companion to know all of the time.

First, create a lorebook by clicking New.

Then, select the lorebook you just created in the Active World(s) for all chats dropdown box. You can have multiple lorebooks active at a time, it can be a good way to organize your information.

Now, you're going to make a new entry for your lorebook.

Make sure your lorebook is selected in the circled drop-down box.

Press the + sign.

Now click the down-arrow to expand your new entry to set it up.

Now we'll fill in the entry.
Give your entry a title, which is just for you to see; your companion will not see it.

Then, add some keywords; they can contain spaces, just as long as they're separated by a comma. Keywords are what you'll type to trigger the entry content to be inserted into the chat context. Companions are not aware of what the keywords are.

And now for the content. This is the exact information your companion will recieve when you use a keyword, make sure it is self-contained. You can make this as long or as short as you want.

For example, I have an entry for my childhood stuffed rabbit Hoppy. Any time I use the keywords in my message, my companion will have information about what I'm talking about.

A useful feature is Sticky, which allows the entry content to stay active in context for the number of messages you set it to. After that number of messages from the user has passed, the entry content will no longer be in context, only what was said directly in the chat about it.

Managing Chats

Now that we have done all of that setup, there should be a nice chat environment set up for you!

You can start typing into the chat bar to begin a new chat with your companion.

If you want to regenerate a message or close a chat, both options can be found in the hamburger button menu on the left of the chat bar. 

Helpful Links
You did it. You fucking did it. You've gotten through setting up SillyTavern. Seriously, pat yourself on the back, it's a really overwhelming interface.

One of the most useful extensions for troubleshooting is Prompt Inspector, which is an easy way to examine your entire context window.

For further help, check out the following links:

- SillyTavern Documentation
- SillyTavern Discord
- SillyTavern Subreddit