Creator is a state management library that enables concise, fluid, readable and testable business logic code.

Read and update state with compile time safety:

// Creator creates a stream of data.
final counter = Creator.value(0);

Widget build(BuildContext context) {
  return Column(
    children: [
      // Watcher will rebuild whenever counter changes.
      Watcher((context, ref, _) => Text('${}')),
        // Update state is easy.
        onPressed: () => context.ref.update<int>(counter, (count) => count + 1),
        child: const Text('+1'),

Write clean and testable business logic:

// repo.dart

// Pretend calling a backend service to get fahrenheit temperature.
Future<int> getFahrenheit(String city) async {
  await Future.delayed(const Duration(milliseconds: 100));
  return 60 + city.hashCode % 20;

// logic.dart

// Simple creators bind to UI.
final cityCreator = Creator.value('London');
final unitCreator = Creator.value('Fahrenheit');

// Write fluid code with methods like map, where, etc.
final fahrenheitCreator = cityCreator.mapAsync(getFahrenheit);

// Combine creators for business logic. 
final temperatureCreator = Emitter<String>((ref, emit) async {
  final f = await;
  final unit =;
  emit(unit == 'Fahrenheit' ? '$f F' : '${f2c(f)} C');

// Fahrenheit to celsius converter.
int f2c(int f) => ((f - 32) * 5 / 9).round();

// main.dart

Widget build(BuildContext context) {
  return Watcher((context, ref, _) => 
      Text( ?? 'loading'));
... context.ref.set(cityCreator, 'Pairs');  // triggers backend call
... context.ref.set(unitCreator, 'Celsius');  // doesn't trigger backend call

Getting started:

dart pub add creator

Table of content:

Why Creator?

When we built our Flutter app, we started with flutter_bloc. Later we switched to riverpod. However, we encountered several issues related to its async providers and realized we want a different mechanism.

So we built Creator. It is heavily inspired by riverpod, but with simpler data model, better async support and much simplified implementation.

The benefit using Creator:

  • Enables concise, fluid, readable and testable code.
  • Expressive for both sync and async behavior.
  • Simple concept. Simple implementation. Lightweight.


Creator’s concept is extremely simple. There are only two types of creators:

  • Creator which creates a stream of T.
  • Emitter which creates a stream of Future<T>.

Here stream is in its logical term, not the Stream class.

Both Creator and Emitter:

  • Can depend on other creators, and update its state when others’ state changes.
  • Are loaded lazily and disposed automatically.

Dependencies form a graph, and it is managed internally by the framework. This is the graph for the weather example above:




Creator takes a function you write to create a state. The function takes a Ref, which provides API to interact with the internal graph.

final number = Creator.value(42);  // Same as Creator((ref) => 42)
final double = Creator((ref) => * 2);

Calling watch adds an edge number -> double to the graph, so double‘s create function will rerun whenever number‘s state changes.

The nice part is that creator comes with methods like map, where, etc. They are similar to those methods in Iterable or Stream. So double can simply be:

final double = => n * 2);

To update the creator’s state, use either set or update:

... ref.set(number, 42);  // No-op if value is the same
... ref.update(number, (n) => n + 10);

You can also read a creator when watch doesn’t make sense, for example, inside a touch event handler.

  onPressed: () => print(,
  child: const Text('Print'));

Creator’s dependency can be dynamic:

final C = Creator((ref) {
  final value =;
  return value >= 0 ? value :;

In this example, A -> C always exists, B -> C may or may not exist. The framework will update the graph properly as dependency changes.


Emitter works very similar to Creator, but it creates Future<T> instead of <T>. The main difference is Creator has valid data to begin with, while Emitter might need to wait for some async work before it yields the first data.

In practice, Emitter is very useful to deal with data from backend services, which is async by nature.

final stockCreator = Creator.value('TSLA');
final priceCreator = Emitter((ref, emit) async {
  final stock =;
  final price = await fetchStockPrice(stock);

Emitter takes a FutureOr<void> Function(Ref ref, void Function(T) emit), where ref allows getting data from the graph, and emit allows pushing data back to the graph. You can emit multiple times.

Existing Stream can be converted to Emitter easily with It works both sync or async:

final authCreator =
    (ref) => FirebaseAuth.instance.authStateChanges());

final userCreator = async {
  final authId = await
      authCreator.where((auth) => auth != null).map((auth) => auth!.uid));
  return FirebaseFirestore.instance.collection('users').doc(authId).snapshots();

This example also shows the extension method where and map. With them, userCreator will only recreate when auth id changes, and ignore changes on other auth properties.

In some sense, you can think Emitter as a different version of Stream, which makes combining streams super easy.

Emitter generates Future<T>, so it can be hooked to Flutter’s FutureBuilder for UI. Or you can use Emitter.asyncData, which is a creator of AsyncData<T>. AsyncData is similar to AsyncSnapshot for future/stream:

enum AsyncDataStatus { waiting, active }

class AsyncData<T> {
  const AsyncData._(this.status,;
  const AsyncData.waiting() : this._(AsyncDataStatus.waiting, null);
  const AsyncData.withData(T data) : this._(, data);

  final AsyncDataStatus status;
  final T? data;

With it building widget with Emitter is easy:

Watcher((context, ref, _) {
  final user =;
  return user != null ? Text(user!.name) : const CircularProgressIndicator();


To make creators work, wrap your app in a CreatorGraph:

void main() {
  runApp(CreatorGraph(child: const MyApp()));

CreatorGraph is a InheritedWidget. It holds a Ref object (which holds the graph) and exposes it through context.ref.

CreatorGraph uses DefaultCreatorObserver by default, which prints logs when creator state changes. It can be replaced with your own log collection observer.


Watcher is a StatefulWidget. It takes builder function Widget Function(BuildContext context, Ref ref, Widget child). You can use ref to watch creators to populate the widget. child can be used optionally if the subtree should not rebuild when dependency changes:

Watcher((context, ref, child) {
  final color =;
  return Container(color: color, child: child);
}, child: ExpensiveAnimation());  // this child is passed into the builder above

Listen to change

Watching a creator will get its latest state. What if you also want previous state? Simply call watch(someCreator.change) to get a Change<T>, which is an object with two properties T? before and T after.

For your convenience, Watcher can also take a listener:

// If builder is null, child widget is directly returned. You can set both
// builder and listener. They are independent of each other.
Watcher(null, listener: (ref) {
  final change =;
  print('Number changed from ${change.before} to ${change.after}');
}, child: SomeChildWidget());


Creators can have names for logging purpose. Setting name is recommended for any serious app.

final numberCreator = Creator.value(0, name: 'number');
final doubleCreator = => n * 2, name: 'double');

Keep alive

By default, creators are disposed when losing all its watchers. This can be overridden with keepAlive parameter. It is useful if the creator maintains a connection to backend (e.g. listen to firestore realtime updates).

final userCreator = {
  return FirebaseFirestore.instance.collection('users').doc('123').snapshots();
}, keepAlive: true);

Creator group

Creator group can generate creators with external parameter. For example, in Instagram app, there might be multiple profile pages on navigation stack, thus we need multiple instance of profileCreator.

// Instagram has four tabs: instagram, reels, video, tagged
final tabCreator = Creator.arg1<Tab, String>((ref, userId) => 'instagram');
final profileCreator = Emitter.arg1<Profile, String>((ref, emit, userId) async {
  final tab =;
  emit(await fetchProfileData(userId, tab));  // Call backend

// Now switching tab in user A's profile page will not affect user B.
... ref.set(tabCreator('userA'), 'reels');

Creators comes with factory methods arg1 arg2 arg3 which take in 1-3 arguments.

Extension method

Our favorite part of the framework is that you can use methods like map, where on creators (full list here). They are similar to those methods in Iterable or Stream.

final numberCreator = Creator.value(0);
final oddCreator = numberCreator.where((n) => n.isOdd);

Note that Creator needs to have valid state at the beginning, while where((n) => n.isOdd) cannot guarantee that. This is why where returns an Emitter rather than a Creator. Here is the implementation of the where method. It is quite simple and you can write similar extensions if you want:

extension CreatorExtension<T> on Creator<T> {
  Emitter<T> where(bool Function(T) test) {
    return Emitter((ref, emit) {
      final value =;
      if (test(value)) {

You can use extension methods in two ways:

// Define oddCreator explicitly as a stable variable.
final oddCreator = numberCreator.where((n) => n.isOdd);
final someCreator = Creator((ref) {
  return 'this is odd: ${}');

// Create "oddCreator" anonymously on the fly.
final someCreator = Creator((ref) {
  return 'this is odd: ${ => n.isOdd))}');

If you use the “on the fly” approach, please read next section about creator equality.

Creator equality

The graph checks whether two creators are equal using ==. This means creator should be defined in global variables, static variables or any other ways which can keep variable stable during its life cycle.

What happens if creators are defined in local variables on the fly?

final text = Creator((ref) {
  final double = Creator((ref) => * 2);
  return 'double: ${}';

Here double is a local variable, it has different instances whenever text is recreated. The internal graph could change from number -> double_A -> text to number -> double_B -> text as the number changes. text still generates correct data, but there is extra cost to swap the node in the graph. Because the change is localized to only one node, the cost can be ignored as long as the create function is simple.

If needed, an optional List<Object?> args can be set to ask the framework to find an existing creator with the same args in the graph. Now when number changes, the graph wont change:

final text = Creator((ref) {
  // args need to be globally unique. ['text', 'double'] is likely unique.
  final double = Creator((ref) => * 2, args: ['text', 'double']);
  return 'double: ${}';

The same applies to using extension methods on the fly:

final text = Creator((ref) {
  return 'double: ${ => n * 2, args: ['text', 'double']))}';

Internally, args powers these features:

  • Creator group. profileCreator('userA') is a creator with args [profileCreator, 'userA'].
  • Async data. userCreator.asyncData is a creator with args [userCreator, 'asyncData'].
  • Change. number.change is a creator with args [number, 'change'].

Service locator

State management libraries are commonly used as service locator:

class UserRepo {
  void changeName(User user, String name) {...}
final userRepo = Creator.value(UserRepo());

..., name);

If needed, ref can be passed to UserRepo Creator((ref) => UserRepo(ref)). This allows UserRepo read or set other creators. Do not watch though, because it will make UserRepo rebuild.

Error handling

The framework will:

  • For Creator, store exception happened during create and throw it when watch.
  • For Emitter, naturally use Future.error, so error is returned when watch.
  • In either case, error is treated as a state change.

This means that errors can be handled in the most natural way, at the place makes the most sense. Use the weather app above as an example:

// Here we don't handle error, meaning it returns Future.error if network error
// occurs. Alternately we can catch network error and return some default value,
// add retry logic, convert network error to our own error class, etc.
final fahrenheitCreator = cityCreator.mapAsync(getFahrenheit);

// Here we choose to handle the error in widget.
Widget build(BuildContext context) {
  return Watcher((context, ref, _) {
    try {
      return Text( ?? 'loading');
    } catch (error) {
      return TextButton('Something went wrong, click to retry', 
          onPressed: () => ref.recreate(fahrenheitCreator));


Testing creator is quite easy by combining watch, read, set. Use the weather app above as an example:

test('temperature creator change unit', () async {
  final ref = Ref();
  expect(await, "60 F");
  ref.set(unitCreator, 'Celsius');
  await Future.delayed(const Duration()); // allow emitter to propagate
  expect(await, "16 C");

test('temperature creator change fahrenheit value', () async {
  final ref = Ref();
  expect(await, "60 F");
  ref.set(fahrenheitCreator, Future.value(90));
  await Future.delayed(const Duration()); // allow emitter to propagate
  expect(await, "90 F");




A counter app shows basic Creator/Watcher usage.



Simple weather app shows splitting backend/logic/ui code and writing logic with Creator and Emitter.



Simple app shows how the creator framework build internal graph dynamically.

Best practice

Creator is quite flexible and doesn’t force a particular style. Best practices also depend on the project and personal preference. Here we just list a few things we follow:

  • Split code into repo files (backend service call), logic files (creator) and UI files (widget).
  • Define creator in global variables.
  • Keep creator small for testability. Put derived state in derived creators (using map, where, etc).

Read source code

Creator’s implementation is surprisingly simple. In fact, the core logic is less than 500 lines of code.

Read creator_core library in this order:

  • graph.dart: a simple implementation of a bi-directed graph using adjacency list. It can automatically delete nodes which become zero out-degree.
  • creator.dart: the CreatorBase class and its two sub classes, Creator and Emitter. Their main job is to recreate state when asked.
  • ref.dart: manages the graph and provides watch, read, set methods to user.
  • extension.dart: implement extension methods map, where, etc.

Read creator library in this order:

  • creator_graph.dart: A simple InheritedWidget which expose Ref through context.
  • watcher.dart: A stateful widget which holds a Creator<Widget> internally.


Is it production ready?

Well, we have been using it in production for our own app (Chooly). However, since it is new to the community, the API might change as we take feedbacks. So the suggestion for now: read the source code and make your own call.

Is it bad to define creator as global variable?

It’s not. Creator itself doesn’t hold state. States are held in Ref (in CreatorGraph). Defining a creator is more like defining a function or a class.

How to watch a property while still accessing the whole object?

final someCreator = Creator((ref) { =>;
  final user =;
  return '${}\'s email is changed';

How does creator’s life cycle work?

  • It is added to the graph when firstly being watched.
  • It can be removed from the graph manually by Ref.dispose.
  • If it has watchers, it is automatically removed from the graph when losing all its watchers, unless keepAlive property is set.

What’s the difference between context.ref vs ref in Creator((ref) => ...)?

They both point to the same internal graph, the only difference is that the first ref’s _owner field is null, while the second ref’s _owner field is the creator itself. This means:

  • It is the same to read, set or update any creators with either ref. The operation is pass to the internal graph.
  • If ref._owner is null, will simply add foo to the graph.
  • If ref._owner is not null, will also add an edge foo -> ref._owner to the graph.

What’s the difference between Creator<Future<T>> vs Emitter<T>?

They are both extended from CreatorBase<Future<T>>, whose state is Future<T>. However, there are two important differences, which make Emitter<T> better for async tasks:

  • Emitter<T> stores T in addition to Future<T>, so that we can log change of T or populate AsyncData<T> properly.
  • Emitter<T> notify its watcher when T is emitted, so its watchers can start their work immediately. Creator<Future<T>> notify its watchers when the future is started, so its watchers are still blocked until the future is finished.

That’s it

Hope you enjoyed reading this doc and will enjoy using Creator. Feedback and contribution are welcome!


View Github