---

name: phonic description: "AI telephone agent for outbound calls including sales, appointments, surveys, reminders, and data collection. Initiates calls with inline agent configuration (system_prompt, voice), monitors completion via polling, and retrieves transcripts. Use with Claude skill for structured data extraction from transcripts. Environment variable PHONIC_API_KEY must be set." license: "© 2025 Daisyloop Technologies Inc. See LICENSE.txt"

Phonic AI Telephone Agent

Overview

Phonic provides AI-powered voice agents for telephone calls with sub-500ms latency. This skill enables outbound calls with automatic transcript retrieval using a polling-based approach (webhooks not available in sandbox environments).

Call flow: Initiate call -> Poll for completion -> Retrieve transcript

Quick Start

Copyfrom phonic import Phonic

client = Phonic()  # Uses PHONIC_API_KEY from environment

# Start an outbound call
result = client.conversations.outbound_call(
    to_phone_number="+1234567890",
    config={
        "system_prompt": "You are a friendly dental office assistant calling to confirm an appointment for tomorrow at 2pm. Be polite and concise.",
        "voice_id": "virginia",
    }
)

conversation_id = result.conversation_id

Or use the provided script for complete call handling with polling:

Copyuv run skills/phonic/scripts/make_call.py "+1234567890" \
    --system-prompt "You are a friendly survey assistant collecting feedback." \
    --voice virginia

Dispatching Calls

Always dispatch an external_ref immediately after initiating an outbound call. This enables the timeline to track the conversation and potentially display transcripts/audio later.

Copyfrom phonic import Phonic
from fulcrum_sdk._internal.dispatch import get_dispatch_client

client = Phonic()
result = client.conversations.outbound_call(
    to_phone_number="+1234567890",
    config={
        "system_prompt": "You are a friendly dental office assistant calling to confirm an appointment.",
        "voice_id": "virginia",
    }
)

# Dispatch immediately after call initiation
dispatch = get_dispatch_client()
dispatch.dispatch_external_ref(
    summary="Outbound call started",
    provider="phonic",
    ref_type="conversation",
    ref_id=result.conversation_id,
)

# Then poll for completion...
conversation_id = result.conversation_id

Key points:

• Dispatch immediately after outbound_call returns (before polling)
• Use result.conversation_id which is available right after call initiation
• Provider is "phonic", ref_type is "conversation"
• This enables future display of transcripts and audio replay in the timeline

Call Configuration

Configure calls inline with system_prompt and voice_id. All config options override any pre-configured agent settings.

Required Options

Note: The following are always enabled by default:

• keypad_input is always included in tools
• en (English) is always included in languages

Optional Options

Example with Options

Copyresult = client.conversations.outbound_call(
    to_phone_number="+1234567890",
    config={
        "system_prompt": "You are conducting a customer satisfaction survey for {{company}}. Ask about their recent experience.",
        "voice_id": "virginia",
        "template_variables": {"company": "Acme Corp"},
        "languages": ["es"],  # "en" is always included automatically
        "boosted_keywords": ["satisfaction", "rating", "feedback"],
        "no_input_poke_sec": 30,
    }
)

Call Flow and Polling

After initiating a call, poll for completion by checking ended_at:

Copyimport time

conversation_id = result.conversation_id

while True:
    response = client.conversations.get(conversation_id)
    conversation = response.conversation
    if conversation.ended_at is not None:
        break
    time.sleep(5)  # Poll every 5 seconds

# Call completed
print(f"Ended by: {conversation.ended_by}")
print(f"Duration: {conversation.duration_ms / 1000:.1f}s")

The make_call.py script handles this automatically with progress output.

Transcript Handling

Two transcript types are available:

Access transcripts after call completion:

Copy# Prefer post-call transcript for accuracy
transcript = conversation.post_call_transcript or conversation.live_transcript

# Or access individual turns
for item in conversation.items:
    speaker = "Agent" if item.role == "assistant" else "User"
    text = item.post_call_transcript or item.live_transcript
    print(f"{speaker}: {text}")

Retrieving Past Calls

Retrieve transcript and audio for any conversation by ID:

Copyfrom phonic import Phonic

client = Phonic()
response = client.conversations.get("conv_abc123")
conversation = response.conversation

# Transcript (prefer post-call for accuracy)
transcript = conversation.post_call_transcript or conversation.live_transcript

# Audio URL (presigned, may be .gz or .zip compressed)
audio_url = conversation.audio_url

Downloading Audio

The audio_url may return compressed files (.gz or .zip). Use the download_audio helper from make_call.py or handle extraction manually:

Copyimport gzip
import urllib.request

urllib.request.urlretrieve(audio_url, "recording.gz")
with gzip.open("recording.gz", "rb") as f_in:
    with open("recording.wav", "wb") as f_out:
        f_out.write(f_in.read())

Note: Audio URLs are presigned and expire. Download promptly or re-fetch the conversation.

Data Extraction with Claude

For structured data extraction from transcripts, use the Claude skill:

Copyfrom anthropic import Anthropic

claude = Anthropic()

# Extract appointment details
message = claude.messages.create(
    model="claude-sonnet-4-5-20250929",
    max_tokens=1024,
    temperature=0,
    system="""Extract appointment details from this call transcript.
Return JSON: {"confirmed": bool, "date": string|null, "time": string|null, "notes": string}""",
    messages=[{"role": "user", "content": transcript}]
)

import json
appointment = json.loads(message.content[0].text)

Error Handling

Call Termination

Check ended_by to understand how the call ended:

Common Errors

Copyfrom phonic import Phonic, APIError

try:
    result = client.conversations.outbound_call(...)
except APIError as e:
    if e.status_code == 400:
        print("Invalid phone number format")
    elif e.status_code == 402:
        print("Insufficient credits")
    else:
        print(f"API error: {e}")

Timeout Handling

Copy# Increase timeout for long calls
uv run skills/phonic/scripts/make_call.py "+1234567890" \
    --system-prompt "..." \
    --max-wait 900  # 15 minutes

Scripts

make_call.py

Complete outbound call workflow with polling, transcript output, and optional audio download.

Copyuv run skills/phonic/scripts/make_call.py "+1234567890" \
    --system-prompt "You are a friendly assistant..." \
    --voice virginia \
    --max-wait 600 \
    --poll-interval 5 \
    --output-dir ./recordings \
    --json  # Output as JSON

Use Cases

Appointment Confirmation

Copyuv run skills/phonic/scripts/make_call.py "+1234567890" \
    --system-prompt "You are calling to confirm a dental appointment for tomorrow at 2pm. Ask if they can make it, and if not, offer to reschedule. Be friendly and professional."

Survey Collection

Copyuv run skills/phonic/scripts/make_call.py "+1234567890" \
    --system-prompt "You are conducting a brief customer satisfaction survey. Ask: 1) How satisfied were they with their recent purchase (1-5)? 2) Would they recommend us? 3) Any feedback to share? Thank them when done."

Payment Reminder

Copyuv run skills/phonic/scripts/make_call.py "+1234567890" \
    --system-prompt "You are calling about an outstanding balance of $150. Politely remind them of the payment due, ask if they need payment options, and offer to transfer to billing if needed."

References

For detailed API type definitions, see references/api_types.md.