Documentation | Issues | Changelog | Funding ♥

Python VoIP

Async VoIP Python library for the AI age.

Sponsors

Usage

To get started, you will need a SIP account. One is usually included with ISP. Check your ISP's documentation or router for details.

You will need a SIP AOR (URI), which looks like this:

sip:USER:PASSWORD@SIP_SERVER

CLI

A simple echo call can be started with:

uvx 'voip[cli]' sip sip:alice:********@sip.example.com echo

Each command supports an optional --dial argument to initiate an outbound call instead of waiting for an inbound one.

To dial a number, say a message, and hang up automatically:

uvx 'voip[cli]' sip sip:alice:********@sip.example.com say sip:+15551234567@sip.example.com "Your package has arrived."

You can also talk to a local agent (needs Ollama):

uvx 'voip[cli]' sip sip:alice:********@sip.example.com agent --initial-prompt "Hi, I am looking for a Mr. Ron, first name Mo?"

MCP

The voip package ships a ready-made Model Context Protocol (MCP) server that exposes tools to make phone calls on your behalf to any MCP client.

{
  "mcpServers": {
    "VoIP": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "voip[mcp]",
        "mcp"
      ],
      "env": {
        "SIP_AOR": "sip:number:password@example.com"
      }
    }
  }
}

Python API

uv add voip[audio,ai,pygments]

Subclass TranscribeCall and override transcription_received to handle results. Pass it as session_class when answering an incoming call:

import asyncio
import dataclasses

from voip.ai import TranscribeCall
from voip.sip.protocol import SIP
from voip.sip.types import SipURI
from voip.sip.transactions import InviteTransaction
from voip.rtp import RealtimeTransportProtocol
from faster_whisper import WhisperModel


@dataclasses.dataclass(kw_only=True, slots=True)
class TranscribingCall(TranscribeCall):
    def transcription_received(self, text) -> None:
        print(text)


class TranscribeInviteTransaction(InviteTransaction):
    def invite_received(self, request) -> None:
        self.ringing()
        self.answer(
            session_class=TranscribingCall,
            stt_model=WhisperModel("kyutai/stt-1b-en_fr-trfs", device="cuda"),
        )


async def main():
    loop = asyncio.get_running_loop()
    _, rtp_protocol = await loop.create_datagram_endpoint(
        RealtimeTransportProtocol,
        local_addr=("0.0.0.0", 0),
    )
    await loop.create_connection(
        lambda: SIP(
            rtp=rtp_protocol,
            aor=SipURI.parse("sip:alice:********@example.com"),
            transaction_class=TranscribeInviteTransaction,
        ),
        host="sip.example.com",
        port=5060,
    )
    await asyncio.Future()


asyncio.run(main())

For raw audio access without transcription, subclass AudioCall and override audio_received(self, audio: np.ndarray) instead.