ai_abstracted 0.1.0

Provider-agnostic generative AI: one set of contracts for text, image, video, speech, sound-effect, and music generation across many providers.

ai_abstracted

One set of contracts for generative AI, across many providers and every medium.

You pick a capability (text, image, video, speech, sound effect, or music), hand a typed request plus credentials to a provider client, and get back bytes with normalized metadata. The same shape works whether the bytes came from Gemini, Claude, OpenAI, Flux, Veo, ElevenLabs, or Suno. Switching providers is a constructor swap, not a rewrite.

It is pure Dart with no Flutter and no dart:io in the library, so it runs on servers, the Dart VM, CLIs, and inside Flutter apps. It never writes files: you decide how to store or cache what comes back.

Install

dart pub add ai_abstracted

import 'package:ai_abstracted/ai_abstracted.dart';

A first look

Every capability has an in-memory fake, so you can see the shape without a key or a network:

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> main() async {
  final TextGenerator generator = FakeTextGenerator(text: 'Hello there.');

  final result = await generator.generateText(
    const TextRequest(prompt: 'Say hello.'),
  );

  print(result.text); // Hello there.
}

Moving to a real provider is the same call with a real client:

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> main() async {
  final credentials = credentialsFromEnv(ProviderId.claude, Platform.environment);
  if (credentials == null) {
    stderr.writeln('Set ANTHROPIC_API_KEY first.');
    return;
  }

  final generator = ClaudeTextClient(credentials: credentials);
  final result = await generator.generateText(
    const TextRequest(
      prompt: 'Write a one-line haiku about Dart.',
      system: 'You reply with a single line.',
    ),
  );

  print(result.text);
}

Media comes back as bytes. You write them where you want:

import 'dart:io';

import 'package:ai_abstracted/ai_abstracted.dart';

Future<void> main() async {
  final images = OpenAiImageClient(
    credentials: const ProviderCredentials(apiKey: 'sk-...'),
  );
  final result = await images.generateImage(
    const ImageRequest(prompt: 'a red bicycle on a white wall', width: 1024, height: 1024),
  );

  await File('bicycle.png').writeAsBytes(result.bytes);
}

Capabilities and providers

Medium	Providers
Text	Gemini, Claude, Mistral, Ollama
Image	Gemini, OpenAI, Flux
Video	Veo (Veo 3 includes audio)
Speech	ElevenLabs
Sound effect	ElevenLabs
Music	Suno (via sunoapi.org)

Each provider has its own page with setup, examples, and tips. See the provider directory.

What you get

• Typed requests and one GenerationResult (bytes, MIME type, kind, metadata) for every medium.
• A typed error hierarchy that maps HTTP failures to AiAuthException, AiRateLimitException, AiInvalidRequestException, and friends.
• Built-in retries with backoff, and polling for the providers that run long jobs (Veo, Flux, Suno).
• An in-memory fake per capability, so your code stays testable without a network or keys.
• A provider registry for choosing a provider at runtime, and an environment credential loader.

Documentation

The full documentation lives in doc/ and is published at https://simonerich.github.io/dart_ai_abstracted/.

Good places to start:

• Installation
• Your first request
• Core concepts
• Provider directory
• Testing with fakes
• Writing your own provider

Example

A runnable example lives in example/main.dart. It uses a fake, so it runs offline:

dart run example/main.dart

Contributing

Contributions are welcome, especially new providers. See CONTRIBUTING.md for the setup, the quality checks, and how the commit messages are formatted. If you add a provider for your own project, a pull request means everyone else can use it too.

License

MIT. See LICENSE.