Writing Style
Beyond personas, Draft has global writing style settings that apply to all transformations. These control baseline behavior like verbosity, tone, and formatting preferences.
Personas vs. Writing Style
Style Settings
Configure these settings in Draft → Settings → Writing Style:
Verbosity
Controls how much detail Draft includes in transformations:
| Level | Behavior |
|---|---|
| Concise | Shortest possible output. Great for Slack, quick replies. |
| Normal | Balanced output. Suitable for most communication. |
| Detailed | Thorough explanations. Good for documentation. |
Tone
Sets the default communication style:
| Tone | Description |
|---|---|
| Direct | Straightforward, no hedging. Gets to the point. |
| Formal | Professional and polished. Good for external comms. |
| Friendly | Warm and approachable. Great for team messages. |
| Casual | Relaxed and conversational. Personal communication. |
Formatting
Default output format preference:
| Format | When to use |
|---|---|
| Prose | Regular paragraphs. Best for emails and documents. |
| Bullets | Bullet points when appropriate. Good for summaries. |
| Mixed | Draft chooses based on content. Most flexible. |
Hard Rules
Hard rules are absolute constraints that Draft will never break, regardless of instruction or persona. These protect important content:
Preserve Quotes
Quoted text is never modified
Never Change Numbers
Numbers, dates, and figures are preserved exactly
English Variant
US or UK spelling conventions
Critical for accuracy
Configuration File
Writing style settings are stored in your Draft configuration file:
Default Configuration
{
"version": "1.0.0",
"writingStyle": {
"verbosity": "normal",
"tone": "direct",
"formatting": "prose"
},
"hardRules": {
"preserveQuotes": true,
"neverChangeNumbers": true,
"englishVariant": "american"
}
}Advanced: Sampling Parameters
For advanced users, Draft exposes the language model sampling parameters. These control the randomness and creativity of outputs:
| Parameter | Default | Effect |
|---|---|---|
| temperature | 0.7 | Lower = more predictable, higher = more creative |
| topP | 0.9 | Nucleus sampling threshold |
| repeatPenalty | 1.1 | Reduces repetition in output |
| maxTokens | 1024 | Maximum output length |
For power users
Per-Action Settings
Different actions (rewrite, summarize, translate, extract) can have their own system prompts. These are configured in the actionContracts section:
"actionContracts": {
"rewrite": {
"systemPrompt": "You are a precise text transformation engine..."
},
"summarize": {
"systemPrompt": "You are a summarization engine..."
},
"translate": {
"systemPrompt": "You are a translation engine..."
},
"extract": {
"systemPrompt": "You are a data extraction engine..."
}
}Advanced feature