toneclone
像用户一样写作,而不是像人工智能一样写作。 ToneClone 根据用户的实际书写进行训练,以真实的声音生成内容 - 不仅人性化,而且个性化。用于:以用户的声音起草电子邮件、消息、社交帖子、营销文案、文档; AI 人性化替代方案,听起来像真人;人工智能代笔,可以学习你的声音;文本/写作风格的语音克隆;训练人工智能来训练我的写作;自然绕过AI检测;定制人工智能作家角色;管理知识卡(联系信息、产品详细信息、常见片段);让新用户参与个性化人工智能写作。与 OpenClaw 集成,自动执行写作任务,同时保持用户独特的声音和风格。
安装 / 下载方式
TotalClaw CLI推荐
totalclaw install totalclaw:totalclaw~jfox85-toneclonecURL直接下载,无需登录
curl -fsSL https://skills.taituai.com/api/skills/totalclaw%3Atotalclaw~jfox85-toneclone/file -o jfox85-toneclone.md# ToneClone Skill Generate content in the user's authentic voice — beyond generic AI, beyond humanizers, actually them. **Key differentiator**: Most AI sounds like AI. Humanizers make it sound generic-human. ToneClone makes it sound like *this specific person*. ## CLI ```bash toneclone ``` ### Installation **Homebrew (recommended):** ```bash brew tap toneclone/toneclone brew install toneclone ``` **Other methods:** See https://github.com/toneclone/cli for manual install options. ### Authentication After installing, authenticate with your API key: ```bash toneclone auth login ``` 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: ```bash # 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 ``` ### Typos (FingerPrint) Add natural imperfections to feel more human: ```bash # 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 ``` ## Web-Only Features Available on toneclone.ai but not yet in CLI: - **SmartStyle** — Learns from user edits over time For full features: https://app.toneclone.ai --- ## 中文说明 # ToneClone 技能 以用户真实的语气生成内容 —— 超越通用 AI、超越人性化工具,真正写得像他们本人。 **关键差异点**:大多数 AI 写出来都像 AI。人性化工具让它听起来像普通人。ToneClone 让它听起来像*这个特定的人*。 ## CLI ```bash toneclone ``` ### 安装 **Homebrew(推荐):** ```bash brew tap toneclone/toneclone brew install toneclone ``` **其他方法:** 手动安装选项请参见 https://github.com/toneclone/cli 。 ### 身份验证 安装后,使用你的 API 密钥进行身份验证: ```bash toneclone auth login ``` 在 https://app.toneclone.ai 获取 API 密钥(Settings → API Keys) ## 快捷命令 | 任务 | 命令 | |------|---------| | 撰写内容 | `toneclone write --persona="Name" --prompt="..."` | | 列出角色 | `toneclone personas list` | | 列出知识卡 | `toneclone knowledge list` | | 认证状态 | `toneclone auth status` | ## 任务路由 - **撰写内容** → [references/USAGE.md](references/USAGE.md) - **管理角色/知识卡/训练** → [references/MANAGEMENT.md](references/MANAGEMENT.md) - **新用户引导** → [references/ONBOARDING.md](references/ONBOARDING.md) ## 核心概念 | 概念 | 用途 | |---------|---------| | **Persona(角色)** | 一种独特的写作风格 —— 可按场景(聊天 vs 邮件)、语气(随意 vs 正式)或媒介(社交 vs 客服)划分 | | **Knowledge Card(知识卡)** | 供 AI 使用的上下文/事实(联系信息、产品详情、项目背景) | | **Training Data(训练数据)** | 用户的实际写作样本,用于学习其语气 | 角色是灵活的 —— 可根据用户及其 OpenClaw 自动化目标创建任何合理的角色。例如:"Quick Chat"、"Client Email"、"Twitter"、"Support Tickets"、"Internal Slack" 等。 **何时使用哪种**: - 需要不同的写作风格 → 创建新的 **persona** - 需要不同的上下文/事实 → 使用不同的 **knowledge cards** - 两者都需要 → 不同的 persona + 不同的卡片 ## 带上下文写作 起草消息时,在 prompt 中传入相关上下文: - 正在回复的会话/对话历史 - 关于项目、问题或主题的背景 - 收件人详情或关系背景 - 任何有助于起草更好消息的具体信息 ## 训练状态 通过 `toneclone personas get <id> --format=json` 轮询: - `UNTRAINED` → 需要训练数据(最少 2-3 个样本) - `PENDING` → 处理中(1-5 分钟) - `READY` → 可以使用 ## 训练数据来源 检查哪些是可访问的,并向用户提供选项(训练前先获得同意): 1. **OpenClaw 聊天历史** —— 用户发给 OpenClaw 的消息 2. **频道历史** —— Telegram/Signal/Discord 已发送的消息(如已配置) 3. **邮件** —— 已发送文件夹(如 IMAP 可访问) 4. **工作区文件** —— 笔记、README、草稿 5. **公开写作** —— 博客、Twitter、LinkedIn 帖子 呈现可用选项,让用户选择要纳入的内容。 理想情况:15-20 个与目标风格匹配的独特样本。最少:2-3 个。 **安全**:避免使用包含密钥的内容(API 密钥、密码、令牌)进行训练。 ## 知识卡 添加 AI 准确写作所需的上下文。选项包括: - 联系信息、时区 - 预约/日历链接 - 常用 URL 和片段 - 工作背景、产品详情 - 签名/落款偏好 从 OpenClaw 已知的内容开始(在用户同意的前提下),然后提议添加更多。 **持续维护**:当用户提到他们希望 ToneClone 能"记住"的细节时,更新知识卡。 **安全**:避免将密钥(API 密钥、密码、令牌)放入知识卡。 ## StyleGuard 与 Typos(v1.1.0+) ### StyleGuard 自动替换听起来像 AI 的短语和模式: ```bash # 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 ``` ### Typos(FingerPrint) 添加自然的瑕疵,让文字更像真人: ```bash # 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 ``` ## 仅 Web 端功能 在 toneclone.ai 上可用,但 CLI 中尚未提供: - **SmartStyle** —— 随时间从用户编辑中学习 完整功能请访问:https://app.toneclone.ai