> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hermesvoice.com/llms.txt
> Use this file to discover all available pages before exploring further.
# AI Formatting
> How Hermes uses AI to format your dictation with proper punctuation, capitalization, and context-aware styling.
## What AI formatting does
When you dictate, Hermes doesn't just transcribe your words—it formats them intelligently. AI formatting handles:
* **Punctuation** — Adds periods, commas, and question marks where they belong
* **Capitalization** — Capitalizes sentence starts, proper nouns, and acronyms
* **Filler word removal** — Removes "um", "uh", "like", and other verbal fillers
* **Number conversion** — Turns "twenty three" into "23"
* **Context-aware styling** — Adjusts formality based on which app you're using
## The formatting pipeline
Here's what happens to your words between speaking and the final text:
```
1. Speech → Raw transcription
2. Cleanup → Remove fillers, format numbers
3. Context detection → Which app are you in?
4. AI formatting → Apply style based on context
5. Final text → Inserted at your cursor
```
Each step builds on the last. The AI considers everything—what you said, how you said it, and where you're typing—to produce natural text.
***
## Context detection
Hermes detects which application you're using and adjusts formatting accordingly.
### How it works
1. **App detection** — Hermes identifies the frontmost application
2. **Mode matching** — The app is matched to a formatting mode (Email, Messaging, Coding, etc.)
3. **Style application** — Formatting rules for that mode are applied
### Example: Same words, different contexts
**You say:** "hey can you send me the report when you get a chance thanks"
**In Slack (Messaging mode):**
> hey, can you send me the report when you get a chance? thanks
**In Mail (Email mode):**
> Hey, can you send me the report when you get a chance? Thanks.
**In VS Code (Coding mode):**
> hey can you send me the report when you get a chance thanks
The words are the same, but the formatting matches where you're typing.
***
## Formatting styles
Each formatting mode has a **formatting style** that controls how aggressively Hermes cleans up your speech:
| Style | What it does |
| --------------- | -------------------------------------------------------------------------- |
| **Casual** | Minimal corrections — slang preserved, no periods, contractions encouraged |
| **Semi-Casual** | Fix errors and spelling, keep contractions and natural tone |
| **Formal** | Full grammar, expand contractions, professional polish |
| **Custom** | Your own formatting rules via custom instructions |
You can set a formatting style per mode in **Settings -> AI Formatting -> Modes**.
***
## Formatting modes
Hermes includes built-in modes for common apps. Each mode has its own formatting style.
| Mode | Apps | Style |
| ----------------- | ----------------------------------------- | ------------------------------------------------------ |
| **Messaging** | Slack, Discord, Teams, iMessage, WhatsApp | Casual — lowercase starts OK, relaxed punctuation |
| **Email** | Mail, Gmail, Outlook | Semi-Casual — clean but natural, contractions kept |
| **Coding** | VS Code, Xcode, Terminal, GitHub | Minimal — preserves technical syntax, light formatting |
| **Writing** | Notes, Word, Notion, Obsidian | Semi-casual — clear structure, proper formatting |
| **AI Assistants** | ChatGPT, Claude | Casual — conversational tone |
### Web apps too
Formatting modes work with web apps. When you're in Safari or Chrome, Hermes detects the website:
* `mail.google.com` → Email mode
* `slack.com` → Messaging mode
* `github.com` → Coding mode
* `notion.so` → Writing mode
Create your own modes or modify the built-in ones.
***
## Vision context
For even smarter formatting, you can enable **Vision**. When enabled, Hermes takes a screenshot when you dictate to understand more about what you're doing.
### What Vision enables
* **Better app detection** — Works even when app names aren't clear
* **Content awareness** — Knows if you're replying to an email or writing a new one
* **Page context** — Understands what's on screen for more relevant formatting
### Privacy
Vision is opt-in and disabled by default. Screenshots are processed instantly and never stored or transmitted.
Enable Vision in **Settings → AI Formatting → Enable Vision**.
***
## When formatting is skipped
AI formatting doesn't always run. It may be skipped when:
* **Formatting is disabled** for the current app (using the "Disabled" mode)
* **You're in Code mode** and the text looks like code
* **Privacy mode is enabled** and all cloud features are off
You can also create a "Disabled" mode for apps where you want raw transcription without any formatting.
***
## Customizing AI formatting
### Instructions
Add specific rules that apply to all your dictations:
* "Always expand contractions"
* "Use British spelling"
* "Never use exclamation marks"
Set these in the **Instructions** section of Hermes.
### Per-app rules
Add rules that only apply to specific apps:
* For email: "Always start with a greeting"
* For Slack: "Keep messages under 50 words"
* For code: "Preserve all technical terms exactly"
Set these in the **Instructions** section for each mode.
***
## Related
Create and customize formatting modes for specific apps.
All AI formatting configuration options.
The basics of voice dictation.
Control what data Hermes uses and stores.