Build an AI Phone Receptionist With OpenClaw (Answer Calls, Book Appointments, Take Messages)
Your business phone rings at 2 AM. Instead of voicemail, an AI receptionist answers, qualifies the caller, books appointments, and texts you a summary. Here's how.
Voicemail is where leads go to die. 80% of callers who reach voicemail hang up and call the next business. An AI receptionist picks up every call, 24/7, and handles it intelligently.
This isn't science fiction. Voice AI has crossed the quality threshold where callers can't tell — or don't care — that they're talking to a machine. What they care about is getting an answer.
What an AI Receptionist Does
Answers Every Call
Ring. Pick up. "Thanks for calling [Business Name], this is our AI assistant. How can I help you?"
No hold music. No "press 1 for..." menus. No voicemail. A conversational agent that listens, understands, and responds — in under 500 milliseconds.
Handles Common Requests
- "What are your hours?" → Answers immediately
- "Do you have availability this week?" → Checks calendar, offers slots
- "I need to reschedule my appointment" → Finds the booking, offers alternatives
- "What's the price for [service]?" → Quotes from your pricing list
- "Can I speak to [name]?" → Transfers the call or takes a message
Qualifies Leads
Before transferring to you, the receptionist gathers:
- Caller's name and contact info
- What they need
- How urgent it is
- Their budget range (if applicable)
- Best time to call them back
You get a text summary within 30 seconds of the call ending: "Sarah Chen called about kitchen renovation. Budget €25-35K. Wants to start in April. Callback requested before 5 PM."
Takes Intelligent Messages
When the caller needs something beyond the agent's scope:
- Captures detailed message (not just "please call back")
- Asks clarifying questions
- Confirms the message back to the caller
- Sends it to you via text, email, or Slack — immediately
The Voice Stack
Building an AI phone system requires three layers:
Layer 1: Telephony
A phone number that receives calls and converts speech to data:
- Twilio — Most popular, global numbers, €1/month per number + €0.01/minute
- Vonage — Similar pricing, strong European presence
- Telnyx — Developer-focused, slightly cheaper
Layer 2: Speech-to-Text + Text-to-Speech
Converting voice to text (for the agent to understand) and text back to voice (for the caller to hear):
- Deepgram — Fast, accurate, good for real-time conversation
- ElevenLabs — Most natural-sounding voices
- OpenAI Whisper + TTS — Good quality, easy to integrate
Layer 3: The Brain (OpenClaw)
Your OpenClaw agent processes the transcribed text, generates a response, and sends it back through TTS. Same agent, same memory, same personality — just a different interface.
Caller → Phone → Twilio → Speech-to-Text → OpenClaw Agent
↓
Caller ← Phone ← Twilio ← Text-to-Speech ← Response
Cost Per Call
| Component | Cost |
|---|---|
| Phone number | €1/month |
| Twilio minutes | €0.01/min inbound |
| Speech-to-text | €0.006/min (Deepgram) |
| Text-to-speech | €0.015/min (ElevenLabs) |
| LLM API (Claude Sonnet) | ~€0.01 per call |
| Total per 3-min call | ~€0.09 |
100 calls/month = €10. A human receptionist costs €2,000-3,000/month.
Even a virtual receptionist service charges €200-500/month for the same volume. An AI receptionist is 95% cheaper.
Quality Threshold: When Does It Work?
Voice AI works well for:
- Structured conversations — booking, FAQs, message-taking
- Clear audio — landline or quiet mobile environments
- Standard accents — major languages and common accents handle well
- Short calls — under 5 minutes of focused interaction
Voice AI struggles with:
- Heavy accents or dialects — accuracy drops in noisy environments
- Emotional callers — complaints and sensitive situations need humans
- Complex negotiations — anything requiring nuance beyond FAQ
- Multi-party calls — can't track multiple speakers well
The solution: handle the 70% of calls that are routine (hours, pricing, booking) and transfer the 30% that need a human.
The Transfer Protocol
When a caller needs a human:
- Agent says "Let me transfer you to [name]. One moment."
- If the person is available → warm transfer with context
- If unavailable → "They're not available right now. I can take a detailed message and have them call you back within [timeframe]. Would that work?"
The caller never feels abandoned. They either get transferred or get a specific callback commitment.
Industries That Benefit Most
Medical/dental practices: Appointment scheduling, prescription refill requests, after-hours triage. A dental practice handling 50 calls/day saves their front desk staff 3 hours daily.
Legal firms: Intake screening, appointment booking, case status updates. Clients calling at 10 PM get a professional response instead of voicemail.
Home services (plumbers, electricians, etc.): Lead capture while the tech is on a job site. Instead of missing calls during installations, every lead gets answered.
Real estate: Property inquiries, viewing bookings, buyer qualification. Works alongside the WhatsApp agent for full multi-channel coverage.
Setting Up: The Minimum Viable Receptionist
Step 1: Get a Phone Number
Sign up for Twilio. Get a local number in your area code. Cost: €1/month.
Step 2: Deploy OpenClaw on ClawPort
Same agent setup as any other channel. Write the receptionist personality:
## Phone Receptionist — [Business Name]
You are answering the business phone. Callers expect a professional, helpful receptionist.
Tone: Warm, efficient, clear. Speak in short sentences — phone conversations need brevity.
Always:
- Greet with the business name
- Ask how you can help
- Confirm details by repeating them back
- Thank the caller before hanging up
Never:
- Rush the caller
- Use jargon or technical terms
- Pretend to be human (if asked, say "I'm the AI assistant for [Business]")
- Handle complaints — always offer to transfer or take a message
Step 3: Connect the Voice Pipeline
Wire Twilio → Deepgram → OpenClaw → ElevenLabs → Twilio. Several open-source projects handle this wiring, or use a managed voice AI platform that integrates with OpenClaw.
Step 4: Test With Real Calls
Call your own number. Test every scenario:
- Simple question (hours, location)
- Appointment booking
- Transfer request
- Message taking
- Edge case (heavy accent, background noise)
Step 5: Go Live Gradually
- Week 1: Forward calls to the AI only after hours
- Week 2: Forward calls to AI when your line is busy
- Week 3: AI answers all calls, transfers to human when needed
The Multilingual Advantage
A human receptionist speaks 1-3 languages. An AI receptionist speaks all of them.
For businesses in multilingual markets (Netherlands, Belgium, Switzerland, Singapore), this alone justifies the investment. A caller speaks German, the agent responds in German. The next caller speaks French, the agent switches automatically.
No hiring multilingual staff. No training. No scheduling. Just works.
Answer every call, in every language, 24/7. Deploy your AI receptionist on ClawPort — €0.09 per call vs €2,000/month for a human. Go live this week.
Ready to deploy your AI agent?
Get started with ClawPort in 60 seconds. No credit card required.
Get Started FreeRelated Articles
How Nonprofits Use AI Agents (Donor Engagement, Volunteer Coordination, and More)
AI agents aren't just for tech companies. Here's how nonprofits use OpenClaw to automate donor outreach, coordinate volunteers, and answer questions — on a nonprofit budget.
OpenClaw Webhooks: Connect Your Agent to Any App
How to use OpenClaw webhooks to trigger external actions — CRM updates, ticket creation, Slack notifications, and more. Includes example payloads and real use cases.
The $40/Hour Knowledge Worker Is Being Replaced by OpenClaw (Here's What That Means)
Research, data entry, scheduling, report generation — the tasks that cost $40-80/hour are being automated for $0.50/hour with AI agents. How to be on the right side of this shift.
OpenClaw vs Zapier vs Make: When to Use AI Agents vs Workflow Automation
Zapier connects apps. Make builds workflows. OpenClaw thinks. They're not competitors — they're different tools for different problems. Here's when to use each.