什么是 ClawCall AI？

ClawCall AI 是专为 Openclaw Skills 生态系统设计的专业通信工具，使开发人员能够自动化真实的电话外呼。它利用 Telnyx 进行电话基础设施建设，并利用 Deepgram 进行高性能语音处理，允许 AI 智能体以类人的流利度处理随访、提醒和复杂的任务导向型对话。

与简单的自动拨号器不同，该技能优先考虑语境和个性。通过提供详细的背景信息，智能体可以充当特定的员工或代表，确保专业且具有同理心的互动。它是弥合数字工作流与传统语音通信渠道之间鸿沟的理想选择。

下载入口:https://github.com/openclaw/skills/tree/main/skills/shreyjindal81/clawcall-ai

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

ClawCall AI 应用场景

• 为诊所、美容院或专业服务调度和确认预约。
• 跟踪未付发票或余额，并提供具体的付款指令。
• 进行多阶段互动，智能体需要拨回电话并提供所请求的信息。
• 发送针对术后护理或服务检查的自动化且个性化的提醒。

1. 用户通过提供目标电话号码（E.164 格式）以及包含个性和任务详情的丰富上下文字符串来触发技能。
2. 该技能通过 Telnyx 发起连接，并使用托管的 ngrok 隧道或公共 WebSocket 建立实时音频流。
3. AI 智能体通过 Deepgram 处理语音转文本，并使用 ElevenLabs 语音合成生成自然响应。
4. 在通话过程中，智能体遵循特定的任务语境，按照指示处理问题并提供细节。
5. 通话结束以后，技能会保存本地录音，并返回完整的文本转录，以便在 Openclaw Skills 环境中进行进一步处理。

ClawCall AI 配置指南

在运行技能之前，请确保您已设置必要的环境变量：TELNYX_API_KEY、TELNYX_CONNECTION_ID、TELNYX_PHONE_NUMBER 和 DEEPGRAM_API_KEY。

首先，安装必要的依赖项：

npm install

要使用托管隧道发起通话，请使用以下命令结构：

node telnyx_voice_agent.js --to "+15551234567" --ngrok --personality "A warm receptionist" --task "Follow up on the appointment for Max."

ClawCall AI 数据架构与分类体系

ClawCall AI 将交互数据组织为转录文本和媒体文件，以便于审计和记录保存：

name: clawcall-ai
description: Run AI-powered outbound phone calls with Telnyx + Deepgram Voice Agent. Use when the user wants real phone outreach (follow-ups, confirmations, reminders, callbacks) with configurable personality, task context, model, and voice.
metadata: {"openclaw": {"emoji": "??", "requires": {"bins": ["node", "npm"], "env": ["TELNYX_API_KEY", "TELNYX_CONNECTION_ID", "TELNYX_PHONE_NUMBER", "DEEPGRAM_API_KEY"]}, "primaryEnv": "TELNYX_API_KEY", "os": ["darwin", "linux"]}}

ClawCall AI - Outbound Calls

Make realistic AI phone calls with natural conversation flow.

IMPORTANT: Providing Context

When invoking this skill, you MUST provide rich, detailed context. The voice agent will NOT assume any information - it only uses what you explicitly provide.

Required Information to Gather:

1. Phone number (E.164 format: +15551234567)
2. Personality - Who is the agent? Be specific:
   • BAD: "a receptionist"
   • GOOD: "Emma, a warm veterinary receptionist at Pawsitive Care Animal Hospital who has worked there for 5 years and knows all the vets by name"
3. Task - What is the call about? Include ALL relevant details:
   • BAD: "follow up about their pets"
   • GOOD: "Follow up with the Hendersons about Max (golden retriever, had knee surgery last week, needs post-op checkup, should be taking Rimadyl twice daily). Also remind them about the three hundred twenty-five dollar balance. Available slots: Wednesday 2pm, Thursday 10am or 4pm."

Context Checklist:

• Names (caller's name, business name, contact's name)
• Dates and times (be specific: "Tuesday January 15th at 3:00 PM")
• Relevant details (appointment type, order number, service details)
• Fallback options if needed (reschedule times, alternative actions)
• Any reference numbers or IDs the agent might need

Prerequisites

Install JavaScript dependencies (one-time):

npm --prefix {baseDir} install

If using --ngrok, NGROK_AUTH_TOKEN must be configured and the ngrok account must be verified. If not using --ngrok, set PUBLIC_WS_URL to a reachable wss://.../telnyx endpoint.

Commands

Basic call:

node {baseDir}/telnyx_voice_agent.js --to "+15551234567" --ngrok r
  --personality "" r
  --task ""

Full example (complex multi-topic call):

node {baseDir}/telnyx_voice_agent.js r
  --to "+15551234567" r
  --ngrok r
  --personality "Emma, a warm and experienced veterinary receptionist at Pawsitive Care Animal Hospital. You've worked there for 5 years and genuinely love animals. You know all the vets by name - Dr. Chen specializes in surgery, Dr. Patel handles general wellness, and Dr. Morrison is the exotic animals expert. You're organized but personable." r
  --task "Call to follow up with the Hendersons about their pets. They have three animals at your clinic: 1) Max, a 7-year-old golden retriever who had knee surgery last week - need to schedule his two-week post-op checkup and confirm he's been taking his pain medication (Rimadyl, twice daily with food). 2) Whiskers, a 12-year-old tabby cat due for her senior blood panel and dental cleaning - Dr. Patel recommended this at her last visit in October. 3) Pickles, their bearded dragon who needs his annual wellness exam. Also remind them that Max's surgery bill of eight hundred fifty dollars has a remaining balance of three hundred twenty-five dollars after insurance. Payment plans are available if needed. If they want to schedule, available slots this week: Wednesday 2pm, Thursday 10am or 4pm, Friday 9am." r
  --greeting "Hi there! This is Emma calling from Pawsitive Care Animal Hospital. Is this the Henderson household?"

Follow-up calls with transcript context:

When calling back after a previous conversation, include the full transcript in the task to maintain continuity. The agent will understand the context and pick up where you left off.

node {baseDir}/telnyx_voice_agent.js r
  --to "+15551234567" r
  --ngrok r
  --personality "Emma, a warm veterinary receptionist at Pawsitive Care. You called earlier and promised to call back with info." r
  --task "You're calling back as promised. Here's the previous transcript:

---PREVIOUS CALL TRANSCRIPT---
Emma: Hi! This is Emma from Pawsitive Care Animal Hospital.
User: Hi, yes.
Emma: I wanted to confirm the email for your payment portal, but I didn't have it handy. Would you like me to call back?
User: Sure.
Emma: Great, I'll call you right back with that info.
---END TRANSCRIPT---

You looked up the email - it's jhenderson@gmail.com. Call back to confirm the email is correct and let them know the payment portal link has been sent." r
  --greeting "Hi! It's Emma again from Pawsitive Care, calling back like I said I would."

This is useful when:

• The agent promised to call back with information
• You need to follow up on a previous conversation
• Continuing a multi-part interaction

Voice Selection

Always use ElevenLabs voices unless user specifies otherwise:

• elevenlabs/rachel - Female (default)
• elevenlabs/adam - Male
• elevenlabs/josh - Male (deeper voice)

Model Selection

Fast default: gpt-4o-mini

Output

The call transcript will be returned, containing the full conversation. Use this to:

• Confirm task completion
• Extract information gathered during the call
• Report back to the user

You must return the full call transcript to the user after the call ends.

At call end, recording lifecycle logs are also emitted:

• Recording URL discovered
• Recording saved to local disk (RECORDINGS_DIR, default ./recordings)
• Recording deleted from the Telnyx portal after successful local save

Notes

• The agent will NEVER assume information not provided
• If asked something it doesn't know, it will offer to hang up and call back
• ngrok tunnel is automatically managed
• Environment variables must be configured in OpenClaw settings
• If a call connects with no audio, check DEEPGRAM_API_KEY validity/entitlement first
• Recordings are enabled by default and persisted locally