Skip to content

SillyTavern.md

Latest commit

 

History

History
115 lines (75 loc) · 4.91 KB

SillyTavern.md

File metadata and controls

115 lines (75 loc) · 4.91 KB
Raw

Connecting SillyTavern to WilmerAI (Out of Date. Will Update Soon)

NOTE: This is very outdated. I've consolidated all the info I had before, but I haven't really messed with a newer version of ST than what came out in summer of 2024 except very briefly, so I need to actually download the thing and see what the new UI looks like. This is old.

This guide provides comprehensive instructions for connecting SillyTavern to a running WilmerAI instance.

Prerequisites

Before you begin, ensure you have successfully installed and are running both WilmerAI and SillyTavern according to their respective setup instructions.

Step 1: Choose a Connection Method

WilmerAI offers two primary ways to connect from SillyTavern: Text Completion and Chat Completion. The Text Completion method is recommended for full compatibility and proper parsing of instructions.

Text Completion (Recommended)

This method requires a specific instruction template to function correctly.

  1. In SillyTavern, navigate to the API Connections menu by clicking the plug icon at the top.

  2. Select Text Completion.

  3. You can connect using either the OpenAI or Ollama API format. Enter the WilmerAI server address into the appropriate field.

    • Option A: OpenAI Compatible Connect as OpenAI Compatible v1/Completions.

    • Option B: Ollama Connect as Ollama api/generate.

Chat Completions (Not Recommended)

If you choose to use the Chat Completion method, be aware that it may not be as reliable.

  1. In the API Connections menu, select Chat Completion.

  2. Connect using the OpenAI format.

  3. After connecting, navigate to the preset settings (the "A" icon), expand the Character Names Behavior section, and set it to Message Content.

  4. Alternatively, you can leave the SillyTavern setting and instead edit your WilmerAI user file to set chatCompleteAddUserAssistant to true. Do not enable both settings simultaneously, as it may confuse the AI.

Step 2: Configure Instruction and Context Templates

This step is critical for the Text Completion method to work correctly.

  1. In SillyTavern, click the "A" icon to access the formatting and presets panel.

  2. You will see two sections: Instruct and Context. Each has an import button.

  3. Locate the template files in the Doc_Resources/SillyTavern/ directory of your WilmerAI project.

  4. Import the InstructTemplate file into the Instruct section.

  5. Important: The WilmerAI Instruct Template is mandatory and must not be modified. WilmerAI relies on this specific format to parse commands. The format uses unique tags to structure the conversation:

    [Beg_Sys]You are an intelligent AI Assistant.[Beg_User]SomeOddCodeGuy: Hey there![Beg_Assistant]Wilmer: Hello!

    "input_sequence": "[Beg_User]",
    "output_sequence": "[Beg_Assistant]",
    "first_output_sequence": "[Beg_Assistant]",
    "last_output_sequence": "",
    "system_sequence_prefix": "[Beg_Sys]",
    "system_sequence_suffix": "",
  1. You may optionally import the provided context template into the Context section, or modify it as needed. Ensure the Enabled checkbox for the Context Template is checked.

Step 3: Adjust Global Settings

These settings should be configured regardless of your chosen connection method.

  1. Context Length: Click the dials icon (samplers) on the far left of the top bar. Find the Context (tokens) setting, check the unlocked checkbox, and drag the slider to its maximum value (e.g., 200,000+). WilmerAI manages the final context sent to the LLM, but it needs the full chat history from SillyTavern to do so effectively.
  2. Streaming: On the same screen, ensure the Stream checkbox is enabled if you want to see the response generated in real-time.
  3. Instruct Mode Settings: Go back to the "A" icon panel. Under "Instruct Mode," enable Include Names and * Force Groups and Personas*.

Step 4: Enabling Conversation Memory

WilmerAI's stateful memory system (including summaries and searchable vector memory) is triggered by a [DiscussionId] tag in the prompt.

  • To activate memory for a conversation, place a unique [DiscussionId] tag within the chat context. The Author's Note section is a good place for this.
  • The content inside the tag serves as the unique identifier for the memory files. For example: [DiscussionId]CharacterName_Conversation1[/DiscussionId].
  • If you place the tag in a character's persona card, every conversation with that character will share the same memory.
  • You can use SillyTavern's variables to create unique IDs easily. For instance: [DiscussionId]{{char}}_2025-09-15[/DiscussionId].

WilmerAI will detect this tag, use it to manage memory for the current session, and remove it from the text before it is sent to the LLM.