deepseek_v4_rolepaly_instruct

Why we included this project

Roleplay and character-driven chat on DeepSeek models usually leaves the model's thinking process to chance. This guide gives you two copy-paste instructions that nudge the hidden chain-of-thought in a specific direction: one keeps the model in character with bracketed inner monologue, the other forces a detached, analytical planning style. You paste the chosen instruction at the end of your first message, and because the model sees the full conversation history, it stays in effect for the whole session. There's also a small Python helper that appends the right marker to your first message, so API integrations can switch modes without hand-editing prompts. The author is upfront that triggering is not guaranteed every time, which is useful to know before you build on it. If you are working on interactive fiction, persona-driven assistants, or any DeepSeek roleplay setup, the worked examples and the mode-switching tips are worth a look.