2.1k
GitHub Stars
8
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill openclaw/skills --skill openclaw-voice-gpt-realtime- _meta.json492 B
- CHANGELOG.md2.4 KB
- index.ts20.3 KB
- openclaw.plugin.json5.8 KB
- package.json1019 B
- README.md9.9 KB
- SKILL.md2.9 KB
- tsconfig.json668 B
Overview
This skill lets your OpenClaw agent place real phone calls using OpenAI's Realtime API for low-latency speech-to-speech interactions. It dials numbers, navigates IVR menus, detects voicemail, and returns structured call results. Setup requires a Twilio account, an OpenAI Realtime key, and a public HTTPS endpoint that tunnels to your plugin server.
How this skill works
The skill uses a single Realtime model to handle both speech recognition and synthesis with ~200–300ms latency, eliminating separate STT/TTS components. It initiates Twilio outbound calls (or accepts inbound when enabled), wraps prompts with safety guardrails, sends DTMF for menu navigation, detects voicemail, and captures call outcomes and optional recordings. CLI commands and config options control dialing, status, and concurrency limits.
When to use it
- Automate routine phone tasks like reservations, appointment booking, and inventory checks.
- Conduct batch outreach where an agent must call multiple numbers and report structured responses.
- Integrate voice-capable assistants into services that require real-time, natural-sounding phone interactions.
- Test IVR flows or validate phone-based customer journeys with autonomous navigation.
Best practices
- Provide explicit call objectives and required deliverables in the instruction (e.g., confirm time, gather price).
- Use a secure public tunnel (ngrok/Cloudflare/Tailscale) and never expose private keys; keep server bound to localhost and tunnel only the webhook.
- Enable debug mode only in controlled environments because recordings and transcripts may contain sensitive data.
- Set reasonable concurrency limits and monitor costs; default is 5 concurrent calls and ~ $0.31/min total.
- Include DTMF instructions in prompts when menu navigation is expected and allow the agent time to listen before speaking.
Example use cases
- Ask the agent to call a restaurant and reserve a table, then return confirmation details and reservation number.
- Verify product availability by calling store numbers and reporting stock status and SKU information.
- Schedule appointments with service providers and extract appointment confirmation and instructions.
- Run automated checks of business hours or phone-based service availability across a list of locations.
- Receive inbound calls (optional) for simple IVR routing or to hand off to the agent for live interactions.
FAQ
You need a Twilio Account SID and Auth Token, a Twilio voice-capable phone number (E.164), an OpenAI API key with Realtime access, and a public HTTPS URL that forwards to the plugin server.
How is voicemail handled?
The agent detects voicemail and will mark the call outcome accordingly instead of attempting to leave a conversational response; recordings and transcripts help verify detection.
What are typical costs?
Approximate cost is $0.31 per minute total (~$0.06 OpenAI input, ~$0.24 OpenAI output, ~$0.014 Twilio). A 5-minute call is about $1.55.