Embeddings
Dartantic supports generating vector embeddings for text using both OpenAI and Gemini providers, as well as a growing list of other providers (see Supported Providers to find out which do what in human English and Provider Capabilities to find out which providers support what in your code).
Embeddings are useful for semantic search, clustering, and building RAG (Retrieval-Augmented Generation) applications.
Basic Embedding Usage
The following example shows how to generate embeddings for a document and a query.
import 'package:dartantic_ai/dartantic_ai.dart';
void main() async {
final agent = Agent('openai');
// Generate a document embedding
final documentText = 'Machine learning is a subset of artificial intelligence.';
final documentEmbedding = await agent.createEmbedding(
documentText,
type: EmbeddingType.document,
);
print('Document embedding: ${documentEmbedding.length} dimensions'); // Output: Document embedding: 1536 dimensions
// Generate a query embedding
final queryText = 'What is machine learning?';
final queryEmbedding = await agent.createEmbedding(
queryText,
type: EmbeddingType.query,
);
print('Query embedding: ${queryEmbedding.length} dimensions'); // Output: Query embedding: 1536 dimensions
}
Embedding Similarity and Search
Use the built-in cosine similarity function to compare embeddings for semantic search:
import 'package:dartantic_ai/dartantic_ai.dart';
void main() async {
final agent = Agent('openai');
// Create embeddings for different texts
final text1 = 'The cat sat on the mat.';
final text2 = 'A cat is sitting on a mat.';
final text3 = 'The weather is sunny today.';
final embedding1 = await agent.createEmbedding(text1, type: EmbeddingType.document);
final embedding2 = await agent.createEmbedding(text2, type: EmbeddingType.document);
final embedding3 = await agent.createEmbedding(text3, type: EmbeddingType.document);
// Calculate similarities
final similarity1vs2 = Agent.cosineSimilarity(embedding1, embedding2);
final similarity1vs3 = Agent.cosineSimilarity(embedding1, embedding3);
print('Similarity between similar texts: ${similarity1vs2.toStringAsFixed(4)}'); // Output: Similarity between similar texts: 0.8934
print('Similarity between different texts: ${similarity1vs3.toStringAsFixed(4)}'); // Output: Similarity between different texts: 0.2156
// Similar texts should have higher similarity
if (similarity1vs2 > similarity1vs3) {
print('✓ Similar texts are more semantically related'); // Output: ✓ Similar texts are more semantically related
}
}
Cross-Provider Embedding Support
Both OpenAI and Gemini providers support embedding generation with consistent APIs. However, the embeddings themselves are NOT compatible between providers. This means that embeddings generated by one provider cannot be directly compared or used interchangeably with embeddings generated by another provider.
import 'package:dartantic_ai/dartantic_ai.dart';
void main() async {
final text = 'Artificial intelligence is transforming technology.';
// OpenAI embeddings
final openaiAgent = Agent('openai');
final openaiEmbedding = await openaiAgent.createEmbedding(text, type: EmbeddingType.document);
print('OpenAI embedding: ${openaiEmbedding.length} dimensions'); // Output: OpenAI embedding: 1536 dimensions
// Gemini embeddings
final geminiAgent = Agent('gemini');
final geminiEmbedding = await geminiAgent.createEmbedding(text, type: EmbeddingType.document);
print('Gemini embedding: ${geminiEmbedding.length} dimensions'); // Output: Gemini embedding: 768 dimensions
// Note: Different providers may have different embedding dimensions
// But the API and similarity calculations work the same way
}
For a complete implementation, see embeddings.dart.