voiceml 0.7.1

Official Python SDK for the VoiceML REST API (Twilio-compatible voice + AMD service from VoiceTel)

📞 VoiceML Python SDK

The official Python client for the VoiceML REST API — Twilio-compatible outbound voice and answering-machine-detection from VoiceTel, with type-safe, async-ready Python.

📚 Table of Contents

• Features
• Installation
• Quickstart
• Authentication
• Resource Reference
• Error Handling
• Async Support
• Pagination
• Migration from twilio-python
• Rate Limits
• Development
• API Documentation
• Contributors
• Sponsors
• License

✨ Features

🛡️ Strongly Typed End-to-End

• Pydantic v2 models for every one of the 81 API operations — request bodies validated before they leave your machine, responses validated when they arrive.
• Autocomplete everywhere. Your IDE knows the shape of every field — Call.sid, Recording.duration, Queue.current_size are all typed.
• Twilio-compatible wire shapes — account_sid, from_number, to_number, status callbacks, pagination envelopes — match what Twilio's Programmable Voice API documents.

⚡ Sync + Async, Same Surface

• Client for blocking calls, AsyncClient for await-based async — identical method names, identical return types.
• Built on httpx — HTTP/2 ready, connection pooling, custom transports if you need them.
• TLS session cache + persistent connections out of the box.

🔁 Production-Grade Transport

• Automatic retry with exponential backoff on 429 / 5xx — honors Retry-After headers.
• Configurable timeouts per client or per call.
• HTTP Basic auth with AccountSid:ApiKey — exactly what the Twilio SDK uses, so existing credentials work unchanged.
• Structured exception hierarchy — RateLimitError, AuthenticationError, NotFoundError, etc. all subclasses of ApiError you can catch broadly or narrowly.

📞 Complete API Coverage

• Calls — originate, fetch, terminate, update + per-call recordings, streams, siprec, transcriptions, notifications, events, and the /Calls/{sid}/Payments lifecycle (Pay TwiML companion).
• Conferences — list, fetch, end conferences, plus participants (mute / hold / kick) and conference-scoped recordings.
• Queues — create, list, update, delete, peek, dequeue (front or specific member).
• Applications — CRUD on stored TwiML + callback bundles.
• Recordings — account-wide list, metadata fetch, audio fetch (follows S3 redirect), delete.
• Messages — create, fetch, list (To/From/DateSent filters + pagination), update (Body redaction; Status=canceled), delete.
• IncomingPhoneNumbers — list, fetch, update.
• Notifications — fetch, list.
• Diagnostics — /health deep probe, OpenAPI spec.

🧪 Tested

• 57 unit tests with mocked HTTP layer (respx) and real Pydantic validation on every fixture — spec drift gets caught at parse time.
• Integration test suite that runs against a callBroadcast / VoiceML instance — gated by env vars, safe for CI.

📦 Clean Distribution

• Zero codegen footprint — every byte hand-written.
• Built with hatchling; ships as wheel + sdist.
• py.typed marker — downstream type checkers see your imports natively.

🚀 Installation

pip install voiceml

Requires Python 3.10 or later. Tested against 3.10, 3.11, 3.12, and 3.13.

🏁 Quickstart

from voiceml import Client
from voiceml.models import CreateCallRequest

with Client(account_sid="AC…", api_key="…") as c:
    call = c.calls.create(
        CreateCallRequest(
            To="[phone]",
            From="[phone]",
            Url="https://example.com/twiml",
            MachineDetection="DetectMessageEnd",
        )
    )
    print(call.sid, call.status)

    for q in c.queues.list().queues:
        print(q.friendly_name, q.current_size)

🔑 Authentication

Every endpoint uses HTTP Basic with your AccountSid as the username and your per-tenant API key as the password — identical to Twilio's auth shape, so credentials issued for Twilio code work here unchanged.

from voiceml import Client

with Client(account_sid="AC…", api_key="…") as c:
    me = c.diagnostics.health()  # uses your AccountSid + key on every call

Don't have credentials yet? See voicetel.com/docs/api/v0.6/voiceml/ for issuance and rotation.

🗺️ Resource Reference

Resource	Sync + Async	Covers
client.calls	originate, fetch, list, terminate, update	+ per-call recordings, streams, siprec, transcriptions, notifications, events, payments
client.conferences	list, fetch, end	participants (mute / hold / kick), conference-scoped recordings
client.queues	create, list, update, delete	peek, dequeue (front or specific member)
client.applications	CRUD on TwiML + callback bundles
client.recordings	account-wide list, metadata, audio fetch, delete	follows S3 redirect for audio
client.messages	create, fetch, list, update, delete	To/From/DateSent filters; Body redaction; Status=canceled
client.incoming_phone_numbers	list, fetch, update
client.notifications	fetch, list
client.diagnostics	/health, OpenAPI spec

Every method that takes a request body accepts a typed Pydantic model imported from voiceml.models:

from voiceml import Client
from voiceml.models import CreateCallRequest, StartPaymentRequest

with Client(account_sid="AC…", api_key="…") as c:
    call = c.calls.create(CreateCallRequest(
        To="[phone]",
        From="[phone]",
        Url="https://example.com/twiml",
    ))
    # On a live call, open a Pay session:
    session = c.calls.start_payment(call.sid, StartPaymentRequest(
        IdempotencyKey="order-482917",
        StatusCallback="https://example.com/pay-status",
    ))
    print(session.sid, session.status)

🚨 Error Handling

All HTTP errors raise subclasses of voiceml.ApiError. Catch broadly or narrowly:

Status	Exception
400	BadRequestError
401	AuthenticationError
403	PermissionDeniedError
404	NotFoundError
409	ConflictError
410	GoneError
429	RateLimitError
501	NotImplementedAPIError
5xx	ServerError
other	ApiError

from voiceml import Client, NotFoundError, RateLimitError

with Client(account_sid="AC…", api_key="…") as c:
    try:
        call = c.calls.get("CA0000000000000000000000000000aaaa")
    except NotFoundError:
        print("That call isn't on your account.")
    except RateLimitError as e:
        print(f"Slow down — retry in {e.body.get('retry_after', '?')}s")

The Twilio-compatible error body (code, message, more_info, status) is parsed into error.code / error.message with the raw payload on error.body.

⚡ Async Support

Identical surface to Client, with await-based methods:

import asyncio
from voiceml import AsyncClient

async def main() -> None:
    async with AsyncClient(account_sid="AC…", api_key="…") as c:
        calls = await c.calls.list(status="in-progress")
        for call in calls.calls:
            print(call.sid, call.duration)

asyncio.run(main())

📄 Pagination

List operations return a …List model with a Twilio-compatible pagination envelope (page, page_size, total, next_page_uri, previous_page_uri, …). For /Calls and /Messages, use the iter() helper to walk all pages transparently:

for call in c.calls.iter(status="completed", page_size=200):
    process(call)

for msg in c.messages.iter(from_number="[phone]", page_size=200):
    archive(msg)

For other resources, page manually with client.<resource>.list(page=n).

🔁 Migration from twilio-python

The account_sid + api_key pair Twilio's SDK validates in its constructor works unchanged here:

# Before — Twilio
from twilio.rest import Client as TwilioClient
client = TwilioClient("AC…", "<token>", region=None)

# After — VoiceML (Twilio-compatible)
from voiceml import Client
client = Client(account_sid="AC…", api_key="<api-key>")

Method names follow the resource map above (client.calls.create(...), client.queues.list(), …) rather than Twilio's client.api.v2010.accounts(sid).calls.create(...) chain — flatter, fewer keystrokes, same wire format on the way out.

⏱️ Rate Limits

VoiceML applies per-tenant rate limits at the edge. The SDK automatically retries 429 responses with Retry-After honored, up to max_retries (default 2). To bump it:

Client(account_sid="AC…", api_key="…", max_retries=4, timeout=60.0)

🛠️ Development

git clone https://github.com/voicetel/voiceml-python-sdk
cd voiceml-python-sdk
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# Unit tests (fast, no network)
pytest tests/unit

# Lint + type-check
ruff check src tests
mypy src

# Integration tests (live, read-only against a configured VoiceML instance)
cp .env.example .env  # fill in VOICEML_ACCOUNT_SID / VOICEML_API_KEY / VOICEML_BASE_URL
pytest tests/integration

# Build wheel + sdist
python -m build
twine check dist/*

📖 API Documentation

• Reference docs: voicetel.com/docs/api/v0.6/voiceml/
• Validator: voicetel.com/voiceml/validator/
• SDK catalogue: voicetel.com/docs/voiceml-sdks/
• Type definitions: see the voiceml.models module — every wire shape has a Pydantic model.

🙌 Contributors

• Michael Mavroudis — Lead Developer

Contributions welcome. Open an issue describing the change you want to make, or send a pull request against main.

💖 Sponsors

Sponsor	Contribution
VoiceTel Communications	Primary development and production hosting

📄 License

MIT with the Commons Clause restriction. See LICENSE and voicetel.com/legal/.