Skip to main content
Out of the box support for 8 built-in providers, with more available via the openai_compat.dart example.

​
Provider Capabilities

ProviderDefault ModelDefault Embedding ModelCapabilitiesNotes
OpenAIgpt-4otext-embedding-3-smallChat, Embeddings, Vision, Tools, StreamingFull feature support
OpenAI Responsesgpt-5text-embedding-3-smallChat, Embeddings, Vision, Tools, Streaming, ThinkingIncludes built-in server-side tools
Anthropicclaude-sonnet-4-0-Chat, Vision, Tools, Streaming, ThinkingExtended thinking with token budgets
Googlegemini-2.5-flashtext-embedding-004Chat, Embeddings, Vision, Tools, Streaming, ThinkingExtended thinking with dynamic budgets
Mistralmistral-large-latestmistral-embedChat, Embeddings, Tools, StreamingEuropean servers
Coherecommand-r-plusembed-english-v3.0Chat, Embeddings, Tools, StreamingRAG-optimized
Ollamallama3.2:latest-Chat, Tools, StreamingLocal models only
OpenRoutergoogle/gemini-2.5-flash-Chat, Vision, Tools, StreamingMulti-model gateway
Note: Together AI, Google-OpenAI, and Ollama-OpenAI have been moved to the openai_compat.dart example. You can still use them by registering them with Agent.providerFactories.

​
Provider Configuration

ProviderProvider PrefixAliasesAPI KeyProvider Type
OpenAIopenai-OPENAI_API_KEYOpenAIProvider
OpenAI Responsesopenai-responses-OPENAI_API_KEYOpenAIResponsesProvider
AnthropicanthropicclaudeANTHROPIC_API_KEYAnthropicProvider
Googlegooglegemini, googleaiGEMINI_API_KEYGoogleProvider
Mistralmistral-MISTRAL_API_KEYMistralProvider
Coherecohere-COHERE_API_KEYCohereProvider
Ollamaollama-None (local)OllamaProvider
OpenRouteropenrouter-OPENROUTER_API_KEYOpenAIProvider

​
Setup

# Set API keys
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."
export GEMINI_API_KEY="..."

​
Usage

// Basic
Agent('openai');

// With model
Agent('anthropic:claude-3-5-sonnet');

// Chat + embeddings
Agent('openai?chat=gpt-4o&embeddings=text-embedding-3-large');

// Extended thinking (OpenAI Responses, Anthropic, Google)
Agent('anthropic:claude-sonnet-4-5', enableThinking: true);

// Server-side tools (OpenAI Responses)
Agent('openai-responses:gpt-5');

​
Find Providers

// All providers
Agent.allProviders

// By name (returns fresh instance)
Agent.createProvider('claude') // → anthropic provider

// For runtime capability discovery, use Provider.listModels()
final provider = Agent.createProvider('openai');
await for (final model in provider.listModels()) {
  print('${model.name}: ${model.kinds}'); // Shows chat, embeddings, media, etc.
}

​
Custom Config

final provider = OpenAIProvider(
  apiKey: 'key',
  baseUrl: Uri.parse('https://custom.api.com/v1'),
);
Agent.forProvider(provider);

​
Custom Headers

All providers support custom HTTP headers for enterprise scenarios like authentication proxies, request tracing, or compliance logging: 
final provider = GoogleProvider(
  apiKey: apiKey,
  headers: {
    'X-Request-ID': requestId,
    'X-Tenant-ID': tenantId,
  },
);

final agent = Agent.forProvider(provider);
Custom headers flow through to all API calls and can override internal headers when needed. This works consistently across OpenAI, Google, Anthropic, Mistral, and Ollama.

​
Check Capabilities

Use Provider.listModels() for runtime capability discovery: 
// List models and their capabilities
final provider = Agent.createProvider('openai');
await for (final model in provider.listModels()) {
  print('${model.name}: ${model.kinds}');
  // Output: gpt-4o: {chat}, text-embedding-3-small: {embeddings}
}

// Check if a specific model supports embeddings
final models = await provider.listModels().toList();
final hasEmbeddings = models.any((m) => m.kinds.contains(ModelKind.embeddings));
if (hasEmbeddings) {
  final agent = Agent('openai');
  final embed = await agent.embedQuery('test');
} else {
  print('Provider does not support embeddings');
}

​
Model Kinds

Models are categorized by their kind via ModelKind:
  • chat - Chat conversations
  • embeddings - Vector embeddings
  • media - Media generation (images, documents, etc.)

​
Typed Output With Tools

The typedOutputWithTools capability indicates a provider can handle both function calling and structured JSON output in the same request:
ProviderSupportImplementation
OpenAI✅Native response_format
OpenAI Responses✅Native response_format
Anthropic✅return_result tool
Google✅Double agent orchestrator
OpenRouter✅OpenAI-compatible
Ollama❌Coming soon
Cohere❌API limitation
Together AI, Google-OpenAI, and Ollama-OpenAI also support typed output with tools when registered via the openai_compat.dart example. Anthropic’s approach uses the return_result tool pattern. This is a special tool that allows the model to return a structured JSON response. This is handled automatically by the Anthropic provider. Google’s approach uses a transparent two-phase workflow: Phase 1 executes tools, Phase 2 requests structured output. This is handled automatically by the Google provider.

​
List Models

// List all models from a provider
final provider = Agent.createProvider('openai');
await for (final model in provider.listModels()) {
  final status = model.stable ? 'stable' : 'preview';
  print('${model.name}: ${model.displayName} [$status]');
}

// Example output:
// - openai:gpt-4-0613  (chat)
// - openai:gpt-4  (chat)
// - openai:gpt-3.5-turbo  (chat)

​
Examples