LLM drop-in for SDTL2 (inspired by SillyTavern RP)
This is a drop-in for an SDTL2 source tree you already run. It does not include SDTL2, the game, the large language model, KoboldCPP, electron, or Flash.
This mod create a sidebar chat interface that works in tandem with existing dialogue functionality. Additionally, the LLM powers her automations via CustomAutomation inspired hooks as well as responding to user chats open-ended. A model on your machine picks one motion pattern and one spoken line. The game plays that pattern and says the line.
Also includes generic command buttons.
What you need
SDTL2: https://www.undertow.club/downloads/sdtl-2.4987/
Custom Automation (copy included as mods/Automode): https://www.undertow.club/downloads/custom-automation.2516/
DialogueActions (copy included): https://www.undertow.club/downloads/dialogue-actions-v4-08.7578/
Weights: inflatebot/MN-12B-Mag-Mell-R1 · Hugging Face
Q5_K_M and the other quants: bartowski/MN-12B-Mag-Mell-R1-GGUF · Hugging Face
KoboldCpp (built on llama.cpp): Releases · LostRuins/koboldcpp
Node.js + libs (installer handles for windows via winget)
What you copy in
- `src/server/routes/llmRoute.ts` and `src/extensions/standard/llm.ts`, then the pastes in `ADD-TO-SDTL.txt` (`Server.ts`, then `extensions.json`).
- `mods/LLMBridge` — speaks the model's line into the dialogue system.
- `mods/DialogueActions` and `mods/Automode` — parse the bracket and run the motion. Skip either one if you already load it.
- `llm-character.txt` — her name, description, and profile. Sits next to `launch.bat`. The motion rules are not in this file.
- `optional/Briar` — the character the prompt is written as. The loop runs on whoever is loaded. Copy her only if you want that sprite.
`NOTES.md` is the install and the ports. `MODDING.txt` is the after-setup list: which file to open when you want to change her voice, the motion, or the sampling.
Run
Double-click `launch.bat` in the SDTL2 folder. That is the only file you need to start.
On a new machine it runs `install.bat` first. That wizard checks for Node.js. If `node` is missing it asks `Install Node.js LTS now? [y/n]`. `y` installs it (winget, or the official installer if winget is not there). If `node_modules` is missing it runs `npm install` once, then returns to `launch.bat`, which builds and opens the game. The next launch sees Node and the dependencies and skips both.
KoboldCpp on `127.0.0.1` port `5001` is what the LLM tab calls. A GGUF that will return JSON has to be loaded there. The filename does not matter. A censored model usually refuses, and the game then pauses.
Credits
- SDTL 0.1.2 by some_dude
- DialogueActions by Pimgd (sby fork) and CustomAutomation by colin, included because the model speaks ordinary dialogue triggers
This is a drop-in for an SDTL2 source tree you already run. It does not include SDTL2, the game, the large language model, KoboldCPP, electron, or Flash.
This mod create a sidebar chat interface that works in tandem with existing dialogue functionality. Additionally, the LLM powers her automations via CustomAutomation inspired hooks as well as responding to user chats open-ended. A model on your machine picks one motion pattern and one spoken line. The game plays that pattern and says the line.
Also includes generic command buttons.
What you need
SDTL2: https://www.undertow.club/downloads/sdtl-2.4987/
Custom Automation (copy included as mods/Automode): https://www.undertow.club/downloads/custom-automation.2516/
DialogueActions (copy included): https://www.undertow.club/downloads/dialogue-actions-v4-08.7578/
Weights: inflatebot/MN-12B-Mag-Mell-R1 · Hugging Face
Q5_K_M and the other quants: bartowski/MN-12B-Mag-Mell-R1-GGUF · Hugging Face
KoboldCpp (built on llama.cpp): Releases · LostRuins/koboldcpp
Node.js + libs (installer handles for windows via winget)
What you copy in
- `src/server/routes/llmRoute.ts` and `src/extensions/standard/llm.ts`, then the pastes in `ADD-TO-SDTL.txt` (`Server.ts`, then `extensions.json`).
- `mods/LLMBridge` — speaks the model's line into the dialogue system.
- `mods/DialogueActions` and `mods/Automode` — parse the bracket and run the motion. Skip either one if you already load it.
- `llm-character.txt` — her name, description, and profile. Sits next to `launch.bat`. The motion rules are not in this file.
- `optional/Briar` — the character the prompt is written as. The loop runs on whoever is loaded. Copy her only if you want that sprite.
`NOTES.md` is the install and the ports. `MODDING.txt` is the after-setup list: which file to open when you want to change her voice, the motion, or the sampling.
Run
Double-click `launch.bat` in the SDTL2 folder. That is the only file you need to start.
On a new machine it runs `install.bat` first. That wizard checks for Node.js. If `node` is missing it asks `Install Node.js LTS now? [y/n]`. `y` installs it (winget, or the official installer if winget is not there). If `node_modules` is missing it runs `npm install` once, then returns to `launch.bat`, which builds and opens the game. The next launch sees Node and the dependencies and skips both.
KoboldCpp on `127.0.0.1` port `5001` is what the LLM tab calls. A GGUF that will return JSON has to be loaded there. The filename does not matter. A censored model usually refuses, and the game then pauses.
Credits
- SDTL 0.1.2 by some_dude
- DialogueActions by Pimgd (sby fork) and CustomAutomation by colin, included because the model speaks ordinary dialogue triggers
