### Installation

**Homebrew (recommended):**

**Other methods:** See https://github.com/toneclone/cli for manual install options.

### Authentication

After installing, authenticate with your API key:

Get an API key at https://app.toneclone.ai (Settings → API Keys)

## Quick Commands

| Task | Command |
|------|---------|
| Write content | `toneclone write --persona="Name" --prompt="..."` |
| List personas | `toneclone personas list` |
| List knowledge | `toneclone knowledge list` |
| Auth status | `toneclone auth status` |

## Task Routing

- **Writing content** → [references/USAGE.md](references/USAGE.md)
- **Managing personas/knowledge/training** → [references/MANAGEMENT.md](references/MANAGEMENT.md)
- **New user onboarding** → [references/ONBOARDING.md](references/ONBOARDING.md)

## Core Concepts

| Concept | Purpose |
|---------|---------|
| **Persona** | A distinct writing style — by context (chat vs email), tone (casual vs formal), or medium (social vs support) |
| **Knowledge Card** | Context/facts for the AI (contact info, product details, project background) |
| **Training Data** | User's actual writing samples to learn their voice |

Personas are flexible — create whatever makes sense for the user and their OpenClaw automation goals. Examples: "Quick Chat", "Client Email", "Twitter", "Support Tickets", "Internal Slack", etc.

**When to use which**:
- Different writing style needed → Create new **persona**
- Different context/facts needed → Use different **knowledge cards**
- Both → Different persona + different cards

## Writing with Context

When drafting messages, pass relevant context in the prompt:
- Thread/conversation history being replied to
- Background on the project, issue, or topic
- Recipient details or relationship context
- Any specifics that help draft a better message

## Training Status

Poll with `toneclone personas get <id> --format=json`:
- `UNTRAINED` → needs training data (minimum 2-3 samples)
- `PENDING` → processing (1-5 min)
- `READY` → good to go

## Training Data Sources

Check what's accessible and offer options to the user (get consent before training):

1. **OpenClaw chat history** — User's messages to OpenClaw
2. **Channel history** — Telegram/Signal/Discord sent messages (if configured)
3. **Email** — Sent folder (if IMAP accessible)
4. **Workspace files** — Notes, READMEs, drafts
5. **Public writing** — Blogs, Twitter, LinkedIn posts

Present available options and let user choose what to include.

Ideal: 15-20 distinct samples matching target style. Minimum: 2-3.

**Security**: Avoid training on content containing secrets (API keys, passwords, tokens).

## Knowledge Cards

Add context the AI needs to write accurately. Options include:
- Contact info, timezone
- Booking/calendar links
- Common URLs and snippets
- Work context, product details
- Signature/sign-off preferences

Start with what OpenClaw already knows (with user consent), then offer to add more.

**Ongoing**: Update cards when user mentions details they wish ToneClone could "remember".

**Security**: Avoid putting secrets (API keys, passwords, tokens) in knowledge cards.

## StyleGuard & Typos (v1.1.0+)

### StyleGuard
Auto-replace AI-sounding phrases and patterns:

### Typos (FingerPrint)
Add natural imperfections to feel more human:

# List current rules
toneclone personas style-guard list <persona>

# Add custom rule (AI rewrites contextually)
toneclone personas style-guard add <persona> --word "utilize" --mode AI

# Add custom rule (fixed replacement)
toneclone personas style-guard add <persona> --word "in order to" --mode CUSTOM --replacement "to"

# Apply curated bundle (limited or comprehensive)
toneclone personas style-guard bundle apply --persona <persona> --type comprehensive

# Check current settings
toneclone personas typos get <persona>

# Enable with intensity preset (none/subtle/noticeable/high)
toneclone personas typos set <persona> --enable --intensity subtle

# Custom rate (0.0-0.02) with protections
toneclone personas typos set <persona> --rate 0.008 --protected urls,emails,code