announcer-sdk

Python SDK for Announcer — send transactional email from your own domain, DKIM-signed.

Sync and async clients, full type hints, one dependency (httpx).

pip install announcer-sdk

Send an email

from announcer import Announcer

announcer = Announcer()  # reads ANNOUNCER_API_KEY

announcer.send(
    from_="Acme <billing@acme.com>",
    to="customer@example.com",
    subject="Your receipt",
    text="Thanks for your order.",
    html="<p>Thanks for your order.</p>",
)

That's the whole integration. from_ has a trailing underscore because from is a Python keyword; if you already have a dict, splat it and use the plain spelling:

announcer.send(**{"from": "billing@acme.com", "to": "customer@example.com", "text": "Hi"})

to, cc and bcc each take one address or a list, and reply_to sets the reply address — see Several recipients.

Async

Same surface, awaited:

from announcer import AsyncAnnouncer

async with AsyncAnnouncer() as announcer:
    await announcer.send(
        from_="billing@acme.com",
        to="customer@example.com",
        subject="Your receipt",
        text="Thanks!",
    )

Before your first send

You need a registered, verified sending domain — Announcer will not let you send From: a domain you have not proved you control.

domain = announcer.domains.create("acme.com")

for record in domain.dns:
    print(record.type, record.name, record.value)
# TXT mail._domainkey.acme.com v=DKIM1; k=rsa; p=MIIBIjANBg...

# Publish that record, wait for DNS, then:
announcer.domains.verify(domain.id)

One TXT record is the entire ask. SPF and MX stay on Announcer's own bounce domain, so your root domain's DNS is untouched.

What the SDK does for you

Retries are safe. Every send carries an Idempotency-Key, generated per call when you do not supply one. A timeout, a 500, or a 429 gets retried with exponential backoff and jitter — and because the key travels with the retry, the API recognises it as the same operation instead of sending twice.

Supply your own key to extend that guarantee across process restarts:

announcer.send(
    from_="billing@acme.com",
    to="customer@example.com",
    subject="Your receipt",
    text="Thanks!",
    idempotency_key=f"receipt-{order.id}",  # this order mails exactly once, ever
)

A replay tells you so rather than pretending it sent again:

result = announcer.send(..., idempotency_key="receipt-4711")
if result.idempotent_replay:
    ...  # already sent earlier; nothing went out a second time

Errors are typed. Catch the case you can actually handle:

from announcer import SuppressedRecipientError, RateLimitError, PermissionDeniedError

try:
    announcer.send(from_=sender, to=recipient, subject=subject, text=body)
except SuppressedRecipientError:
    # They hard-bounced or complained before. Don't retry; mark them inactive.
    deactivate(recipient)
except RateLimitError as exc:
    print(f"Slow down for {exc.retry_after}s")
except PermissionDeniedError:
    # Domain not registered, not verified, or this key is send-scoped.
    ...

The full set: ValidationError (with a per-field .errors dict), AuthenticationError, PermissionDeniedError, NotFoundError, ConflictError, UnprocessableEntityError, SuppressedRecipientError, RateLimitError, InternalServerError, APIConnectionError, APITimeoutError. All subclass AnnouncerError.

Results are dataclasses, not dicts. Timestamps arrive as datetime objects, date fields as date. The API mixes camelCase and snake_case depending on the endpoint; the SDK normalises everything to Python's convention and derives the booleans you actually want (domain.verified, key.revoked, endpoint.disabled). Free-form payload and detail dicts pass through untouched — those keys are your data.

Attachments

Pass the file's bytes; the SDK does the base64.

from pathlib import Path
from announcer import Attachment

announcer.send(
    from_="billing@acme.com",
    to="customer@example.com",
    subject="Invoice 1042",
    text="Your invoice is attached.",
    attachments=[Attachment("invoice-1042.pdf", Path("invoice.pdf").read_bytes())],
)

At most 20 files and 10 MB per message, text and HTML included. Add a content_id to show an image inline with <img src="cid:logo">. Attachments are sent and never stored, and executables are refused.

Several recipients

to, cc and bcc each take one address or a list. Everything in to and cc is one email whose recipients see each other; bcc recipients see nobody, not even each other:

announcer.send(
    from_="billing@acme.com",
    to=["customer@example.com", "partner@example.com"],
    cc="accounting@acme.com",
    bcc="archive@acme.com",
    reply_to="support@acme.com",
    subject="Your receipt",
    text="Thanks!",
)

At most 50 addresses across the three. reply_to is a header only — it costs nothing and cannot bounce.

Recipients are the billable unit. That call counts four against your quota, not one. It is also what keeps monthly_hard_cap meaningful: otherwise a leaked key could send fifty times your ceiling by padding the list.

One email, or many?

For anything list-shaped — a newsletter, a digest, a fan-out — you want send_many, not a list:

results = announcer.emails.send_many(
    ["a@example.com", "b@example.com", "c@example.com"],
    from_="news@acme.com",
    subject="September update",
    html=body,
)

for r in results:
    if not r.ok:
        print(f"{r.to} failed: {r.error}")

Suppressed recipients

A recipient on your suppression list is dropped and the rest still goes out:

result = announcer.send(
    from_="billing@acme.com",
    to=["good@example.com", "bounced-before@example.com"],
    subject="Your receipt",
    text="Thanks!",
)

result.recipients  # 1 — what actually went out and what you were billed
result.suppressed  # ['bounced-before@example.com']

SuppressedRecipientError is raised only when every recipient is suppressed (or every to recipient — a message with no visible primary recipient is refused rather than sent). Its .suppressed list names them all.

Webhooks

Register an endpoint, store the secret, verify every delivery:

endpoint = announcer.webhooks.create("https://acme.com/hooks/announcer")
print(endpoint.secret)  # whsec_... — shown once, store it now

Flask:

import os
from flask import Flask, request
from announcer import verify_webhook, SignatureVerificationError

app = Flask(__name__)

@app.post("/hooks/announcer")
def announcer_webhook():
    try:
        event = verify_webhook(
            request.get_data(),  # raw bytes, NOT request.get_json()
            request.headers.get("X-Announcer-Signature"),
            os.environ["ANNOUNCER_WEBHOOK_SECRET"],
        )
    except SignatureVerificationError:
        return "", 400

    if event.event == "delivered":
        mark_delivered(event.message.id)
    elif event.event in ("bounced", "complained"):
        deactivate(event.message.to)

    return "", 200

Django is the same with request.body.

Verification checks the HMAC and the timestamp, rejecting anything more than five minutes old so a captured delivery cannot be replayed at you. Tune it with tolerance=.

Events: sent, delivered, bounced, complained, suppressed.

API reference

Client

Announcer(api_key=None, *, base_url=None, timeout=30.0, max_retries=2,
          user_agent=None, headers=None, http_client=None)
AsyncAnnouncer(...)  # same arguments

Methods

domains.*, api_keys.* and webhooks.* need a full-scoped key. Everything else works with a send key too — give integrations send.

Scopes

Mint a send-scoped key for anything that only sends mail:

key = announcer.api_keys.create("production-worker", "send")

A leaked send key cannot register domains, mint successor keys, or touch billing. It is the difference between an incident and a catastrophe.

Contributing

pip install -e ".[dev]"
pytest        # no network; everything runs against httpx.MockTransport
mypy

License

MIT