Node 22+, PostgreSQL, and Redis. Migrations ship in the package — point drizzle-kit at node_modules/notifkit/drizzle — and in development notifkit starts throwaway Postgres and Redis containers for you, so Docker is the only prerequisite to try it.

What it does

Shipping one notification is easy. Shipping a notification system is not: you end up with queues, retry logic, a preference store, quiet-hours maths, provider adapters, and a dead-letter queue nobody wants to own.

notifkit is that machinery, running on your servers. You describe who needs to know what; it works out how — the right channel, at a decent hour, only to people who said yes.

await notifkit.notify({
  user: "usr_123",
  template: "order-shipped",
  channels: ["push", "email"],
  fallback: true, // push first; email only if push fails
});

That call returns immediately. Behind it: preference and consent filtering, timezone-aware quiet hours, per-user throttling, template rendering, provider dispatch with retries and circuit breakers, and a durable log of what happened.


Channels	`email`, `sms`, `push`, `webhook`
Targeting	A user, a list of users, a segment, or a topic
Priorities	`low`, `normal`, `high`, `critical` — separate stream lanes
Scheduling	Future sends with `sendAt`, quiet-hours deferral, cancel before dispatch
Preferences	Per-user channel and topic opt-outs, quiet hours, contact-level overrides
Workflows	Multi-step sequences with `wait`, `waitForEvent`, and `notify` steps
Reliability	Redis Streams, 24h idempotency, backoff retries, DLQ, provider circuit breakers
Templates	`{{var}}` interpolation with escaping decided by the destination field
AI	Optional LLM augmentation before render, via the Vercel AI SDK
Multi-tenancy	Projects with isolated keys, data, and rate limits
Consent	RFC 8058 one-click unsubscribe; complaints and hard bounces suppress automatically
Reporting	Tag a send with a `campaign` label, then read delivery and engagement totals back
Agents	An MCP server (`@notifkit/mcp`) driving all of it from a terminal
Observability	Prometheus `/metrics`, `/health`, `/live`, `/ready`, and a queryable delivery log

Bring your own provider accounts — first-party packages ship for Resend and Firebase Cloud Messaging, and anything else is one Transport class with a send() method. Your keys, your billing, your deliverability.

Requirements: Node 22+, PostgreSQL, Redis. In development notifkit starts throwaway Postgres and Redis containers for you, so the only prerequisite to try it is Docker.

What it is not: a marketing automation suite. It is infrastructure for product notifications.

Documentation

Everything lives at notifkit.dev/docs.


Quickstart	Install to first delivered notification, about ten minutes
How it works	The five nouns and the pipeline they move through
Channels & fallback	Multicast, ordered fallback, writing a transport
Preferences & quiet hours	The four gates every notification passes
Templates & AI	Interpolation, escaping, per-channel content
Segments & scheduling	Fan-out, priority lanes, `sendAt`, idempotency
Workflows	Multi-step sequences, recurring sends, digests
Examples	The five runnable projects in `examples/`
Architecture	Streams, delivery guarantees, topologies, data model
Deployment	Dockerfile, Compose, splitting API from workers
Operations	Health, metrics, the DLQ, key rotation, shutdown
Reference	Every endpoint, payload shape, and SDK method
MCP server	Send and report on campaigns from a terminal agent

Star the repo

If notifkit saves you a month or two you were about to spend building this yourself, give it a star — it is the cheapest way to help other people find it.

Contributing

Issues and pull requests are welcome. npm install && npm run build, then npm test — the suite starts its own Postgres and Redis containers, so Docker is the only thing you need running.

License

MIT. Do what you like with it, including commercially. See LICENSE.