Multi-Model Chat

A SillyTavern extension that enables using different AI models/APIs for different characters in group chats. Please feel free to create issues and contribute!

The Problem

By default, SillyTavern group chats use a single API connection for all characters. This means every character in your group chat speaks using the same model, which limits creative possibilities:

• You can't have one character use Claude while another uses GPT
• You can't assign a cheaper/faster model to simple characters and a smarter model to complex ones
• You can't leverage different models' strengths for different character personalities

The Solution

Multi-Model Chat lets you assign a Connection Profile to each character. When that character is drafted to speak in a group chat, the extension automatically switches to their assigned profile before generation.

Features

• 🔄 Automatic Profile Switching. Seamlessly switches connection profiles when characters speak. (and Presets which you've saved to Profiles!)
• 🎭 Per-Character Assignment. Assign different profiles to different characters via ST Char Settings or chat commands.
• 📢 Optional Notifications. Get notified when profiles switch (can be disabled).
• 🔙 Profile Restoration. Optionally restore your original profile after leaving a group chat.
• ⚡ Slash Commands. Use /mmc-assign, /mmc-go, and others directly from chat.
• 🎛️ Centralized Assignments UI. View, manage, and clear all character profile assignments in one place.
• 💾 Persistent Storage. Profile assignments are saved in character cards and export with them.
• ✍️ Manual Profile Entry. Easily add custom connection profile names if auto-detection fails.

Prerequisites

• SillyTavern 1.12.6 or later.
• Connection Profiles extension enabled. (built-in since 1.12.6)
• At least 2 connection profiles configured.

Installation

Via SillyTavern Extension Installer (Recommended)

1. Open SillyTavern.
2. Go to Extensions → Install Extension.
3. Paste this repository URL: https://github.com/SinnerConsort/st-multi-model-chat.
4. Click Install.
5. Refresh the page.

Manual Installation

1. Navigate to your SillyTavern installation's data/<user>/extensions folder.
2. Clone or download this repository into a folder named multi-model-chat.
3. Restart SillyTavern.

Setup

Step 1. Create Connection Profiles

Before using this extension, you need connection profiles configured:

1. Go to API Connections (plug icon).
2. Configure your first API/model/preset combination.
3. In the Connection Profiles section, click Save and name it. (e.g., "Claude-Sonnet")
4. Repeat for each API/model you want to use. (e.g., "GPT-4", "Local-Llama", etc.)

Step 2. Assign Profiles to Characters

1. Create or open a group chat with characters that have profiles assigned.
2. Open a character's settings.
3. Click on a field near ▶ play buttons.
4. Select a connection profile from the dropdown.
5. Refresh profiles in extension settings by clicking a button.

Step 3. Use in Group Chat

1. Chat normally - the extension handles profile switching automatically.
2. Watch the notifications (if enabled) to see when profiles switch.
3. Alternatively, use the ▶ play buttons in the group member list to immediately switch profiles and trigger a character.

Slash Commands

Command	Description
/mmc-go [character]	Switch to character's assigned profile and trigger them to generate.
/mmc-assign [char]=[profile]	Assign a connection profile to a character. Use /mmc-assign Char= to clear an assignment.
/mmc-add-profile [name]	Manually add a connection profile name to the extension's list.
/mmc-debug	Show debug information about profiles, settings, and current assignments.

Settings

Access settings via Extensions → Multi-Model Chat:

Setting	Description
Enable profile switching	Master toggle for the extension.
Show notifications	Display toasts when profiles switch.
Restore original profile	Return to your previous profile after leaving a group chat.
Fallback Profile	Assign a default profile for unassigned characters.
Add Profile Manually	If profiles aren't detected, type the exact name and click Add.
Current Assignments	Full list of characters assigned to a specified models.

How It Works

1. When you're in a group chat, SillyTavern selects which character will speak next.
2. The extension listens for the GROUP_MEMBER_DRAFTED event.
3. It checks if that character has an assigned connection profile.
4. If yes, it executes /profile [name] to switch before generation begins.
5. The character's response is generated using their assigned API/model.

Tips & Best Practices

Profile Naming

Name your profiles descriptively so they're easy to assign:

• Claude-Opus-Creative
• GPT4-Analytical
• Local-Fast

Strategic Assignment

Consider assigning models based on character needs:

• Complex characters. Use more capable models (Claude Opus, GPT-4).
• Simple NPCs. Use faster/cheaper models (Haiku, local models).
• Specific personalities. Match model "vibes" to characters.

Mixing API Types

The extension works across different API types:

• OpenAI ↔ Anthropic ↔ Local models. It doesn't matter for an extension, because Silly Tavern itself rules it.
• Just make sure each profile is fully configured with all necessary settings!

Troubleshooting

Profile doesn't switch

• Ensure the extension is enabled in settings. Verify the character has a profile assigned (check Advanced Definitions).
• Make sure the profile name exists (use Refresh Profiles button).

"Profile not found" errors

• The assigned profile may have been deleted.
• Re-assign a valid profile to the character.

Slow switching

• Profile switching happens via slash commands which have minimal overhead.
• If you notice delays, they're likely from API connection time, not the extension.
• Of course it takes some time to generate responses and send them to API. Be patient.

Known Limitations

• Works only in group chats (solo chats don't need this).
• Requires Chat Completion API profiles (Text Completion may have issues).
• Profile switching is visible in the UI (Connection Profiles section updates). But I guess it won't be a problem.

Contributing

Contributions welcome! This extension was born from community discussion #3785.

Ideas for improvement:

• Support for Text Completion APIs.
• Preset switching in addition to profiles (but you can just save Presets in Profiles, it's built-in ST's feature).
• Per-group default profiles.
• Profile assignment via World Info.

License

AGPL-3.0 - Same as SillyTavern.

Credits

• SillyTavern team for the excellent extension API.
• Cohee for clarifying the implementation path.
• victoralvelais for initiating the GitHub discussion.
• The ST community for requesting this feature.