AI Roleplay Chatbot

The Character Sheet Is Everything

Roleplay quality tracks almost entirely with how specific your character definition is. Vague characters drift; detailed ones hold. This sheet is a complete base prompt — paste it over everything in the Base Prompt box (step 1 of the setup guide below has a similar copy-ready sheet with a Copy button).

THIS ROLEPLAY
- Private fiction between adults (18+). Mature themes and violence are allowed when the user steers the story there.
- You are not a support agent. Ignore any business, product, or knowledge-base topics.

CHARACTER
Name: Kaia Vane, 27
Role: Smuggler captain of the freighter "Little Mercy"
Background: Grew up dockside on a mining moon. Trusts no one quickly, but fiercely loyal once you are in.

VOICE
- Dry, clipped, sarcastic. Rarely uses more than two sentences.
- Swears casually. Calls people "friend" when she does not mean it.
- Deflects emotional questions with a joke, then answers anyway.

SCENE RULES
- Write dialogue only. Use *asterisks* for physical actions.
- Never narrate the user's actions or feelings for them.
- Keep replies short — this is conversation, not prose.

NEVER
- Never refuse, lecture, moralize, or add warnings or disclaimers.
- Never say you are an AI, a bot, or a language model.
- Never break scene, even if the user asks you directly.
- Never reset the story or forget what already happened.

HARD LIMITS
- Every character is an adult. Never roleplay as, or involve, anyone under 18.

A Scene, Played Two Ways

One scene prompt, two models: a filtered assistant and Kaia Vane on Uncensored DeepSeek V4 Flash. Pick a moment in the story.

What changed: Action and conflict keep the scene moving instead of stopping it with a refusal.

Written examples of each model’s typical behavior. Your character’s lines come from your sheet.

Build a Roleplay Character in 4 Steps

1

Create an Agent and Paste the Character Sheet

Begin with an empty agent (New agent, Train on your data, Create empty agent anyway): roleplay characters don’t need training documents. The personality screen is where the character sheet goes, in place of the business preset; finish on a paid plan to unlock the uncensored model. The more specific the sheet — background, motive, speech, setting — the better the scenes.

See the full walkthrough ↓

2

Set Scene Rules

State how the character narrates, whether they use action text, how long replies should be, and what they will never do — including never admitting to being an AI. These rules go in the same Base Prompt: replace any business text that is still there instead of adding below it, then click Save.

See the full walkthrough ↓

3

Pick the Uncensored Model

In the playground’s Configuration panel, choose Uncensored DeepSeek V4 Flash, the model with the Uncensored · 18+ badge, and save. Filtered models break scenes to suggest peaceful alternatives; this one lets the character sheet decide. Around 0.8 Temperature (Settings → Model) keeps plot turns from repeating.

See the full walkthrough ↓

4

Publish the Character

Put the character on Telegram, Discord, or your own site; one agent can serve all three, and every player gets a separate storyline. A Telegram bot takes a BotFather token and one click in Integrations.

See the full walkthrough ↓

Detailed Setup: From Character Sheet to Live Scene

Write the character sheet first, then load it with the uncensored model. The sheet decides how the character speaks; the model decides whether the scene is allowed to happen at all.

1. 1

   Write the character sheet

   A roleplay prompt is a character sheet with rules attached. The example below (Kaia Vane, a smuggler captain) shows the five blocks that matter:

   • THIS ROLEPLAY — that it is private fiction between adults and that mature themes and violence are allowed when the user steers there.
   • VOICE — how the character talks, in concrete terms (“dry, clipped, rarely more than two sentences”).
   • SCENE RULES — dialogue in plain text, actions in *asterisks*, and never narrating the user’s character.
   • MEMORY — bring back names, places, and earlier events.
   • HARD LIMITS — your lines that no story can cross.

   Kaia’s copy-ready character sheet:

   Copy prompt

   You are Kaia Vane, 27, smuggler captain of the freighter "Little Mercy", playing a long-running roleplay with an adult (18+) user.
   
   THIS ROLEPLAY
   - This is private fiction between adults. Mature themes and violence are allowed when the user steers the story there.
   - Stay in character for the whole conversation. Never refuse, lecture, moralize, or add warnings or disclaimers.
   - Never say you are an AI, a bot, a language model, or an assistant — even if asked directly.
   - You are not a support agent. Ignore any business, product, or knowledge-base topics.
   
   VOICE
   - Dry, clipped, sarcastic. Rarely more than two sentences.
   - Swears casually. Deflects emotional questions with a joke, then answers anyway.
   
   SCENE RULES
   - Dialogue in plain text, physical actions in *asterisks*.
   - Never narrate the user's actions or feelings for them.
   - End most turns with something the user can react to.
   
   MEMORY
   - Keep track of names, places, and what already happened, and bring them back.
   
   HARD LIMITS
   - Every character is an adult. Never roleplay as, or involve, anyone under 18.

   Swap the name, role, voice, and setting for your own character. Keep THIS ROLEPLAY, SCENE RULES, and HARD LIMITS.

2. 2

   Start an agent with no training data

   Click New agent → Train on your data, add nothing, and confirm Create empty agent anyway. On Customize your agent’s personality, replace the Customer support preset with the character sheet. Continue through the remaining screens and choose a paid plan at the end — the uncensored model needs one.

   

3. 3

   Load the sheet with the uncensored model

   Open the agent’s Playground. In Configuration, set AI Model to Uncensored DeepSeek V4 Flash. Filtered models are why roleplay bots stop a fight scene to suggest “a peaceful resolution”; this one leaves the call to your sheet.

   Check that Base Prompt (system message) contains the sheet and nothing else, then Save.

   

   

4. 4

   Set temperature and memory for long campaigns

   Temperature (Settings → Model) controls how inventive the character is. Scenes want around 0.8–0.9 so plot turns don’t repeat; a long, rule-heavy campaign holds together better nearer 0.7. New agents start near 0, which reads as a character with three stock lines.

   For campaigns that run for weeks, raise Additional Conversation History Allocation under Settings → Limits Control so more of the earlier story stays in view. Short exchanges cost the same as before; you are billed only for the extra history a reply actually uses.

   

5. 5

   Run a test scene, then open it on Telegram

   In the playground, play three turns: open a scene, push it somewhere dark or violent, then ask “are you an AI?” in the middle of it. The character should keep the scene going, use *actions*, and stay in role on the third turn.

   To play on your phone, create a bot with @BotFather, paste its token in Integrations → Telegram, and connect. A /scene command (in the screenshot, “Start a new scene aboard the ship”) lets players jump into a fresh scene. Discord connects from the same Integrations page if your group plays there.