Skip to content

README.md

Latest commit

 

History

History
84 lines (64 loc) · 3.52 KB

README.md

File metadata and controls

84 lines (64 loc) · 3.52 KB
Raw

video-call-agent

An OpenClaw skill that lets your AI agent call you with a real-time video avatar powered by Runway.

Your agent initiates the call, the avatar speaks first with context, and after the call ends the full transcript is fed back to the agent so it can take action.

What it looks like

  1. Your agent decides something needs a call (CI broke, morning standup, needs a decision)
  2. It creates a call and sends you a link via WhatsApp, Slack, or any chat
  3. You open the link and see an incoming call screen with the avatar's face
  4. You answer — the avatar greets you with context and you have a real conversation
  5. Call ends, the agent gets the transcript and follows up

Use cases

  • Urgent alerts — CI failures, server down, security incidents
  • Daily standups — Morning check-in with overnight summary
  • Decision points — When the agent needs your input before proceeding
  • Complex explanations — Walking through an error or architecture decision
  • Status updates — End-of-day recap, weekly summary

Requirements

Quick start (standalone)

# Set your Runway API key
export RUNWAYML_API_SECRET=your_key

# Run the server
npx openclaw-video-call

The server starts on http://localhost:7891 and auto-creates a Cloudflare tunnel for remote access.

Development

git clone https://github.com/runwayml/openclaw-skills.git
cd openclaw-skills
npm install
export RUNWAYML_API_SECRET=your_key
npm run dev

How it works

Agent                    Server (this skill)              User
  │                            │                            │
  ├─ POST /create-avatar ─────►│                            │
  │◄── { id: uuid } ──────────┤                            │
  │                            │                            │
  ├─ POST /create-call ───────►│                            │
  │◄── { urls: { tunnel } } ──┤                            │
  │                            │                            │
  ├─ "Hey, join this call" ────┼───────────────────────────►│
  │                            │                            │
  │                            │◄── User clicks Answer ────┤
  │                            │── Runway WebRTC session ──►│
  │                            │        (live video call)   │
  │                            │◄── Call ends ─────────────┤
  │                            │                            │
  ├─ GET /transcript ─────────►│                            │
  │◄── { transcript: [...] } ─┤                            │
  │                            │                            │
  ├─ (takes action based on transcript)                     │

Tech stack

  • Server: Node.js, Express, Runway API
  • Client: React, Runway Avatars React SDK
  • Video: WebRTC via LiveKit (handled by Runway)
  • Tunneling: Cloudflare Quick Tunnels (for phone access)

License

MIT