Exa AI

MIT License pub version

Developer Blog • Sapid Labs • GitHub Repo • Pub.dev

Exa API Dart SDK

A pure Dart SDK for the Exa API. Search the web with neural and keyword search, get content from any URL, find similar links, and get AI-powered answers.

Installation

Add this to your package's pubspec.yaml file:

dependencies:
  exa_api: ^1.0.0

Setup

  1. Get your API key from Exa Dashboard
  2. Set up authentication (choose one method):

Set the EXA_API_KEY environment variable at compile time:

# For dart run
dart run --define=EXA_API_KEY=your_api_key_here your_app.dart

# For flutter run
flutter run --dart-define=EXA_API_KEY=your_api_key_here

# Or set as system environment variable
export EXA_API_KEY="your_api_key_here"
dart run your_app.dart

Then initialize the client:

import 'package:exa_api/exa_api.dart';

final exa = ExaApi(
  debugMode: true, // Optional: enable debug logging
);

Option 2: Direct API Key

Initialize the client with an explicit API key:

import 'package:exa_api/exa_api.dart';

final exa = ExaApi(
  apiKey: 'YOUR_EXA_API_KEY',
  debugMode: true, // Optional: enable debug logging
);

API Methods

Perform semantic or keyword searches across the web.

// Basic search
final results = await exa.search(
  query: 'Latest developments in machine learning',
  numResults: 10,
);

// Advanced search with filters
final advancedResults = await exa.search(
  query: 'machine learning research papers',
  type: SearchType.neural,
  category: SearchCategory.researchPaper,
  numResults: 5,
  includeDomains: ['arxiv.org', 'paperswithcode.com'],
  startPublishedDate: DateTime(2023, 1, 1),
  endPublishedDate: DateTime(2023, 12, 31),
  includeText: ['large language model'],
  excludeText: ['survey'],
  contents: const ContentsConfig(
    text: true,
    highlights: HighlightsConfig(
      numSentences: 2,
      highlightsPerUrl: 1,
    ),
  ),
);

print('Found ${advancedResults.results.length} results');
for (final result in advancedResults.results) {
  print('${result.title}: ${result.url}');
}

2. Search and Contents

Convenience method that combines search with content extraction.

final contentResults = await exa.searchAndContents(
  query: 'AI safety research',
  type: SearchType.neural,
  category: SearchCategory.researchPaper,
  numResults: 3,
  text: true,
  highlights: const HighlightsConfig(
    numSentences: 2,
    highlightsPerUrl: 2,
    query: 'Key findings',
  ),
  summary: const SummaryConfig(query: 'Main conclusions'),
  maxCharacters: 1000,
);

for (final result in contentResults.results) {
  print('Title: ${result.title}');
  print('Author: ${result.author ?? 'Unknown'}');
  if (result.text != null) {
    print('Text: ${result.text!.substring(0, 200)}...');
  }
  if (result.highlights != null) {
    print('Highlights: ${result.highlights!.join(' | ')}');
  }
  if (result.summary != null) {
    print('Summary: ${result.summary}');
  }
}

3. Get Contents

Extract content, summaries, and metadata from specific URLs.

final contents = await exa.getContents(
  urls: [
    'https://arxiv.org/abs/2301.00001',
    'https://paperswithcode.com/paper/example',
  ],
  text: true,
  textConfig: const TextConfig(
    maxCharacters: 2000,
    includeHtmlTags: false,
  ),
  highlights: const HighlightsConfig(
    numSentences: 3,
    highlightsPerUrl: 2,
    query: 'methodology',
  ),
  summary: const SummaryConfig(
    query: 'What are the main contributions?',
  ),
  livecrawl: LivecrawlOption.fallback,
  subpages: 2,
  extras: const ExtrasConfig(
    links: 5,
    imageLinks: 3,
  ),
);

for (final content in contents.results) {
  print('URL: ${content.url}');
  print('Title: ${content.title}');
  if (content.text != null) {
    print('Content: ${content.text!.substring(0, 300)}...');
  }
  if (content.summary != null) {
    print('Summary: ${content.summary}');
  }
  if (content.extras?.links != null) {
    print('Links found: ${content.extras!.links!.length}');
  }
}

4. Get Contents Simple

Simplified content extraction with common options.

final simpleContents = await exa.getContentsSimple(
  urls: ['https://example.com/article'],
  text: true,
  highlights: const HighlightsConfig(numSentences: 2),
  summary: const SummaryConfig(),
  maxCharacters: 1500,
  includeHtmlTags: false,
  livecrawl: LivecrawlOption.preferred,
  subpages: 1,
);

final result = simpleContents.results.first;
print('Title: ${result.title}');
print('Content: ${result.text ?? 'No content available'}');

5. Answer

Get AI-powered answers to questions informed by Exa search results.

final answer = await exa.answer(
  query: 'What are the latest breakthroughs in quantum computing?',
  text: true,
);

print('Answer: ${answer.answer}');
print('Source count: ${answer.results.length}');

for (final source in answer.results) {
  print('Source: ${source.title} - ${source.url}');
  if (source.text != null) {
    print('Context: ${source.text!.substring(0, 200)}...');
  }
}

6. Find Similar

Find pages similar to a given URL.

final similarResults = await exa.findSimilar(
  url: 'https://arxiv.org/abs/2301.00001',
  numResults: 5,
  includeDomains: ['arxiv.org', 'openreview.net'],
  startPublishedDate: DateTime(2023, 1, 1),
  includeText: ['transformer', 'attention'],
  contents: const ContentsConfig(
    text: true,
    highlights: HighlightsConfig(
      numSentences: 1,
      highlightsPerUrl: 1,
    ),
    summary: SummaryConfig(query: 'Key innovations'),
  ),
);

print('Found ${similarResults.results.length} similar papers');
for (final result in similarResults.results) {
  print('${result.title}: ${result.url}');
  print('Similarity score: ${result.score}');
  if (result.summary != null) {
    print('Summary: ${result.summary}');
  }
}

Configuration Options

Search Types

SearchType.neural    // AI-powered semantic search (default for most queries)
SearchType.keyword   // Traditional keyword search
SearchType.auto      // Automatically choose the best search type

Search Categories

SearchCategory.researchPaper
SearchCategory.tweet
SearchCategory.movie
SearchCategory.song
SearchCategory.personalSite
SearchCategory.company
SearchCategory.pdfDocument

Livecrawl Options

LivecrawlOption.never      // Only use cached content
LivecrawlOption.fallback   // Use live crawl if not cached (default)
LivecrawlOption.always     // Always perform live crawl
LivecrawlOption.preferred  // Prefer live crawl when possible

Content Configuration

// Text configuration
const textConfig = TextConfig(
  maxCharacters: 2000,
  includeHtmlTags: false,
);

// Highlights configuration
const highlights = HighlightsConfig(
  numSentences: 3,
  highlightsPerUrl: 2,
  query: 'important findings',
);

// Summary configuration
const summary = SummaryConfig(
  query: 'What are the main points?',
);

// Extras configuration (links, images)
const extras = ExtrasConfig(
  links: 10,
  imageLinks: 5,
);

// Complete contents configuration
const contents = ContentsConfig(
  text: true,
  textConfig: textConfig,
  highlights: highlights,
  summary: summary,
  extras: extras,
);

Error Handling

The SDK throws ExaApiException for API errors:

try {
  final results = await exa.search(query: 'test query');
  // Process results...
} on ExaApiException catch (e) {
  print('API Error: ${e.message}');
  if (e.statusCode != null) {
    print('Status Code: ${e.statusCode}');
  }
} catch (e) {
  print('Unexpected error: $e');
}

Cost Management

Monitor API usage costs:

final results = await exa.search(query: 'expensive query');

if (results.costDollars != null) {
  print('Search cost: \$${results.costDollars!.search}');
  print('Contents cost: \$${results.costDollars!.contents}');
  print('Total cost: \$${results.costDollars!.total}');
}

Debug Mode

Enable debug mode to see request/response details:

final exa = ExaApi(
  apiKey: 'YOUR_API_KEY',
  debugMode: true, // Logs requests and responses
);

Examples

See the example file for more comprehensive usage examples.

License

This project is licensed under the MIT License.

Libraries

exa_api