molroo docsSpeaking Style
Give your persona a consistent speaking style extracted from real text.
Speaking Style
molroo can capture and reproduce a character's unique speaking style. Extract a style profile from sample text, set it on your persona, and every response will follow that style — automatically adjusted by the persona's current emotional state.
How it works
Sample text (chat logs, writing samples)
|
v
persona.extractStyleProfile() SDK → API (extraction + storage)
|
v
persona.chat() Style constraints auto-injected into system promptStyle constraints are generated server-side and included automatically in the system prompt. You never need to write style instructions yourself.
Quick start
import { Molroo } from '@molroo-io/sdk';
import { createOpenAI } from '@ai-sdk/openai';
const molroo = new Molroo({ apiKey: 'mk_live_...' });
const openai = createOpenAI({ apiKey: process.env.OPENAI_API_KEY! });
const persona = await molroo.createPersona(
{
identity: { name: 'Sera', role: 'barista' },
personality: { O: 0.7, C: 0.6, E: 0.8, A: 0.9, N: 0.3, H: 0.8 },
},
{ llm: openai('gpt-4o-mini') },
);
// Extract style from sample text — done once
const profile = await persona.extractStyleProfile([
'omg that latte art is gorgeous!!',
'haha yeah i always double-shot my espressos~',
'u should try the oat milk version tho, its rly good',
// ... more messages (20-30 recommended)
]);
// Every chat() call now follows this style
let history: Message[] = [];
const result = await persona.chat('What do you recommend?', { history });
// Response text will naturally match the extracted speaking styleWhat the profile captures
The extracted StyleProfile analyzes patterns in the sample text:
| Category | Examples |
|---|---|
| Vocabulary | Word choice frequency, casual vs. formal register |
| Structure | Sentence length, paragraph patterns, response length |
| Punctuation | Emoji usage, ellipsis patterns, exclamation frequency |
| Register | Formality level, slang density, hedging patterns |
The more sample text you provide, the more accurate the profile. We recommend at least 20-30 messages for reliable extraction.
Emotion-aware modulation
The style isn't static. When the persona's emotional state changes, the speaking style adapts naturally:
- A normally casual character becomes more terse when angry
- Punctuation patterns shift under stress
- Response length may change based on arousal level
This happens automatically — you don't need to manage it.
Extraction options
const profile = await persona.extractStyleProfile(messages, {
timestamps: [1700000000, ...], // Optional: message timestamps for temporal analysis
otherMessages: ['...'], // Optional: other people's messages for contrast
});