SillyTavern Prompt Manager: main prompt, order and toggles
Updated
The Prompt Manager is the list at the bottom of AI Response Configuration when SillyTavern uses Chat Completion. Each row is one block of the request: Main Prompt, card fields, World Info, examples, Chat History and Post-History Instructions. Drag rows to reorder them, switch them off, edit them with the pencil, or add your own blocks at a chosen role and in-chat depth.
Try it now, free, no account neededStart a story with Grimm the Dungeon MasterA full tabletop campaign, run just for youWhere it lives and what the rows are
Open AI Response Configuration (the sliders icon) with a Chat Completion connection and scroll to the bottom. The list there is the recipe for every request, top to bottom. A default preset shows roughly these rows:
- Main Prompt: your standing instructions.
- World Info (before) and World Info (after): lorebook entries, placed around the card.
- Persona Description: who you are.
- Char Description, Char Personality, Scenario: the card's fields.
- Enhance Definitions and Auxiliary Prompt: optional extra instructions, often empty or off.
- Chat Examples: the card's example dialogue.
- Chat History: the conversation itself.
- Post-History Instructions: a block placed after the conversation, read last before the reply.
Each row shows how many tokens it costs, which makes it easy to spot the card or lorebook eating your context.
A Main Prompt and post-history pair that works
The default Main Prompt is one line telling the model to write {{char}}'s next reply. That is a fine start, but most roleplayers replace it. Something like this, pasted with the pencil icon, gives the model a job instead of a role description:
"You are {{char}} in an ongoing roleplay with {{user}}. Write only {{char}}'s words and actions, in third person past tense. React to what {{user}} just did, add one new detail or complication, and stop where {{user}} has something to answer."
Post-History Instructions sit after the chat, so the model reads them right before it writes. Keep them short, because they weigh more than anything else in the prompt:
"[Stay in {{char}}'s voice. Two to four paragraphs. Never write {{user}}'s dialogue or actions.]"
A long rulebook in this slot tends to make every reply sound the same. Put the details in the Main Prompt and keep only the two or three rules that keep slipping here.
Order and toggles
Order matters because models pay most attention to the start and the very end of a request. Leave Chat History near the bottom with Post-History Instructions after it, and put the things that must never be forgotten, such as the character's core, near the top. Moving World Info (after) to just above Chat History is a common tweak when lore keeps getting ignored.
The toggle on each row removes that block from the request without deleting its text. Turn off anything you are not using rather than leaving empty or half-written blocks in place, and test one change at a time so you know which one did something.
Adding your own blocks
The add button opens a form for a new prompt: a name, the text, a role (system, user or assistant) and a position. Relative means it sits wherever you drag it in the list. In-chat means it is inserted inside the conversation at a depth you choose, counted in messages from the end, much like an author's note.
In-chat at depth 2 to 4 is the right home for things that change as the story moves, for example:
"[Scene: the inn's common room, past midnight. The storm has cut the road. Mara still has not said who was in room five.]"
Update it as the scene shifts. Chat depth explains how the counting works and why a shallow depth hits harder.
Cards that bring their own prompts, and saving
A Character Card V2 can carry its own system prompt and post-history instructions. SillyTavern has settings that decide whether those replace yours; when they do, writing {{original}} inside the card's text pulls your own version back in at that spot. If a newly imported character suddenly writes in a different style, check its Advanced Definitions before blaming the model.
Everything in the Prompt Manager is stored in the current Chat Completion preset, so save the preset after editing or the next switch throws your changes away. The utility prompts above the list, such as the impersonation and continue prompts, are saved the same way.
With a UnsavedAI key, the request this list builds reaches Gemma 4 as written, and the Setup preset leaves your prompt blocks untouched when you import it.
Questions people ask
Why can't I see the Prompt Manager?
It only appears when API Connections is set to Chat Completion. Under Text Completion, the equivalent settings live in Advanced Formatting instead.
Is Post-History Instructions the same as a jailbreak?
It is the same slot, renamed. Older guides call the block placed after the chat history the jailbreak prompt.
Should I put the character's personality in the Main Prompt?
No. The card's fields already have their own rows. Keep the Main Prompt about how to write, so it works for every character you load.
