Quick Start

Zero to hero; no PhD required.

Install

dependencies:
  dartantic_ai: ^VERSION

API Keys

Set environment variables for the providers you'll use:

# Native providers
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."
export GEMINI_API_KEY="..."
export MISTRAL_API_KEY="..."

# OpenAI-compatible
export OPENROUTER_API_KEY="..."
export TOGETHER_API_KEY="..."
export COHERE_API_KEY="..."
export LAMBDA_API_KEY="..."

# Ollama runs locally - no key needed

Prefer to handle API keys server-side. Take special care to not expose them in your client code.

See Providers for the complete list.

Your First Chat

The following shows simple agent usage using the Agent() constructor, which takes a model string.

void main() async {
  // Create an agent with a model string
  final agent = Agent(
    'openai',  // Can also use 'openai:gpt-4o' or 'openai/gpt-4o'
  );

  // Run the agent with a prompt and system message
  final result = await agent.send(
    'Where does "hello world" come from?',
    history: [
      const ChatMessage.system('Be concise, reply with one sentence.'),
    ],
  );
  print(result.output); // Output: one sentence on the origin of "hello world"
}

Alternatively, you can use the Agent.forProvider() constructor which takes a provider directly:

void main() async {
  // Create an agent with a provider
  final agent = Agent.forProvider(
    Providers.openai,
  );

  // Run the agent with a prompt and system message
  final result = await agent.send(
    'Where does "hello world" come from?',
    history: [
      const ChatMessage.system('Be concise, reply with one sentence.'),
    ],
  );
  print(result.output); // Output: one sentence on the origin of "hello world"
}

Model Strings

The model string used by Agent can be specified in several ways:

• Just the providerName, e.g. openai → specifies whatever the default chat and embeddings models are for that provider
• providerName:chatModelName, e.g. google:gemini-2.0-flash → specifies a provider and model, separated by a colon
• providerName/chatModelName, e.g. google/gemini-2.0-flash → specifies a provider and model, separated by a slash
• providerName?chat=chatModelName&embeddings=embeddingsModelName → openai?chat=gpt-4&embeddings=text-embedding-ada-002 → uses query format

Providers

A provider is a Dart type like GoogleProvider that knows how to expose model objects. When you pass in a model string, you're really looking up a provider object and one or more types of model objects for the agent to do its work. All the providers support a chat model (e.g. GoogleChatModel) and several of them also support an embeddings model (e.g. GoogleEmbeddingsModel).

You can see the list of Supported Providers that Dartantic exposes. And you can build your own Custom Providers if you like.

Provider Aliases

Some providers support multiple provider prefixes. The alias can be used where the provider name is specified, e.g. gemini:gemini-2.0-flash → google:gemini-2.0-flash. Here's where you can see the list of aliases for each of the Supported Providers.

What's Next?

Now that you've tasted power, dive deeper:

• Examples - Pretty much all the things
• Providers - Pick your AI overlord
• Tools - Give your AI superpowers