SillyTavern setup for a coding assistant

This tutorial adapts SillyTavern's character chat interface into a coding assistant and ChatGPT alternative. The speaker uses an Ada character with a description that says she is comfortable programming in Python, and reports that it produced functional Gradio code. Separate character cards can hold different language or style specifications so users can save and switch between them.

The settings walkthrough recommends increasing response length for code, suggesting that users could go beyond 10,000 tokens. The speaker also recommends lowering temperature to reduce randomness. The demonstration connects SillyTavern to OpenRouter for model selection and discusses using DeepSeek 3.2. This is a hosted model workflow; the tutorial does not demonstrate how to run models locally.

The final configuration step changes the system prompt in Advanced Formatting. The speaker attributes unwanted fabrication to the default prompt's description of a fictional chat. Options shown include a blank prompt, assistant presets and an edited neutral chat preset with the word "fictional" removed, saved as "real chat." That explanation is the speaker's interpretation, rather than proof that a prompt edit prevents hallucinations. The tutorial closes with a brief discussion of extensions and the possibility of writing custom ones.