sigtake

English · Español

Official Python SDK for Sigtake. Sends alerts and signal readings. Sync and async, one dependency, typed.

pip install sigtake

Requires Python 3.10 or newer.

Quickstart

from sigtake import Sigtake

with Sigtake(api_key="sk_...") as sigtake:
    sigtake.alerts.ingest(
        title="Checkout latency above threshold",
        severity="high",
        source="checkout-api",
        team_code="OPS",
        payload={"p95_ms": 2400, "region": "eu-west-1"},
    )

Without api_key, the client reads SIGTAKE_API_KEY from the environment.

Getting an API key

In the app: API Keys → New Key (requires admin or super_admin). The key is shown once — copy it then.

The key resolves both your tenant and your project. There is no project id to pass: one key per project, and staging and production keys are different keys.

The team_code comes from Teams, printed on each team card. It decides who gets notified.

Async

Same surface, with await. Use it in FastAPI or any asyncio app: the sync client would block the event loop on every alert.

from sigtake import AsyncSigtake

async with AsyncSigtake() as sigtake:
    await sigtake.alerts.ingest(title="Payment declined", source="checkout", team_code="PAY")

Create one client per process and reuse it — that is what makes the connection pool worth having.

Alerts

result = sigtake.alerts.ingest(
    title="Payment webhook failing",
    source="payments-worker",
    team_code="PAY",
    severity="critical",  # 'critical' | 'high' | 'medium' | 'low' | 'info', defaults to 'info'
    payload={"attempt": 3},  # optional, must serialize to 8 KB or less
)

result.data.id  # alert id
result.meta.is_duplicate  # True when it folded into an open incident
result.meta.occurrence_count  # how many times this incident has fired

The response never echoes payload back. On a dedup hit the alert you get is the pre-existing one, and its payload may have been written by a different sender.

Alerts are deduplicated on source | title | severity per project. An open alert with the same fingerprint gets its occurrence count bumped instead of opening a second incident, and notifications go out at occurrences 1, 10, 25, 50, 100, then every 100.

team_code must exist in your tenant and the team must be assigned to the project this API key belongs to. If it isn't, the call fails with code == "TEAM_NOT_IN_PROJECT" — assign the team to the project in the app.

Signals

# Flat dict — the common case
sigtake.signals.send("billing-service", {"emails_sent": 42, "queue_depth": 3})

# Explicit readings
sigtake.signals.ingest(
    source="billing-service",
    readings=[{"metric": "emails_sent", "value": 42}],
)

source is the monitor's source key, which must already exist in the project. Up to 100 readings per call, 50 distinct metric names per monitor.

Hitting the metric cap answers with a partial delivery rather than an error — the readings that fit were stored:

result = sigtake.signals.send("billing-service", metrics)
if result.rejected:
    logger.warning("metric cap reached, dropped: %s", result.rejected)

A total rejection (nothing stored) raises SigtakeValidationError.

Errors

Every failure is a SigtakeError subclass carrying status, code and the raw body.

from sigtake import SigtakeRateLimitError, SigtakeValidationError

try:
    sigtake.alerts.ingest(title=..., source=..., team_code=...)
except SigtakeValidationError as err:
    print(err.field_errors)
except SigtakeRateLimitError:
    print("retries exhausted")

408, 429, 5xx and network failures are retried with exponential backoff and jitter. Validation errors never are.

No idempotency. The API does not accept an idempotency key yet, and timeouts are retried. If a request times out after reaching the server, the retry bumps occurrence_count on the existing alert. Nothing is lost or duplicated as an incident, but a notification threshold may fire slightly early.

Configuration

Sigtake(
    api_key="sk_...",  # or SIGTAKE_API_KEY
    base_url="https://api.sigtake.com",  # default; or SIGTAKE_BASE_URL
    timeout=10.0,  # seconds, per attempt
    max_retries=3,  # 0 disables retries
    headers={},  # merged into every request
    http_client=None,  # bring your own httpx.Client; if you pass it, you close it
)

Per call:

sigtake.alerts.ingest(title=..., source=..., team_code=..., timeout=2.0)

Alert ingestion is best treated as fire-and-forget — never let it break the code path that produced the alert:

try:
    sigtake.alerts.ingest(title=..., source=..., team_code=...)
except SigtakeError:
    logger.warning("sigtake ingest failed", exc_info=True)

Development

uv sync --all-groups
uv run ruff check . && uv run mypy
uv run pytest tests/unit --cov=sigtake

Contract tests need a mock served from the spec:

docker run --rm -p 4010:4010 -v "$PWD/spec:/spec" stoplight/prism:5 \
  mock -h 0.0.0.0 --errors -m false /spec/openapi.yaml
uv run pytest tests/contract

More detail in docs/ARCHITECTURE.md.

Releasing

git tag sdk-python-v0.1.0 && git push --tags

The tag triggers the publish workflow, which refuses to publish if the tag and pyproject.toml disagree.

License

MIT