Flutter AI Bridge

A modular, production-ready Flutter package for unified AI provider integration — supports OpenAI, Google Gemini, Anthropic Claude, and local Ollama models with smart routing, agentic function calling, conversation persistence, cost control, RAG utilities, and ready-made UI widgets.

✨ Features

📦 Package Architecture

flutter_ai_bridge/
├── packages/
│   ├── ai_bridge_core         # Core models, interfaces, router, errors, RAG
│   ├── ai_bridge_openai       # OpenAI GPT + Whisper + TTS
│   ├── ai_bridge_gemini       # Google Gemini (with functionDeclarations)
│   ├── ai_bridge_claude       # Anthropic Claude
│   ├── ai_bridge_ollama       # Local Ollama (Edge AI)
│   ├── ai_bridge_ui           # Flutter chat widgets
│   ├── ai_bridge_storage_shared_prefs  # SharedPreferences persistence
│   └── flutter_ai_bridge      # Umbrella package (re-exports core + ui)

🚀 Quick Start

import 'package:flutter_ai_bridge/flutter_ai_bridge.dart';
import 'package:ai_bridge_openai/ai_bridge_openai.dart';
import 'package:ai_bridge_gemini/ai_bridge_gemini.dart';

// 1. Create providers
final openai = OpenAIProvider(
  config: AIConfig(apiKey: 'sk-...', model: 'gpt-4o'),
);
final gemini = GeminiProvider(
  config: AIConfig(apiKey: 'AI...', model: 'gemini-2.0-flash'),
);

// 2. Create the bridge with smart routing
final bridge = AIBridge(
  providers: [gemini, openai],
  strategy: RoutingStrategy.costOptimized,
  budget: TokenBudget(maxTokensPerDay: 100000),
);

// 3. Send a message
final response = await bridge.complete('Hello, AI!');
print(response.content);    // "Hello! How can I help?"
print(response.provider);   // "Gemini" (cheapest first)
print(response.usage);      // AIUsage(prompt: 5, completion: 12, total: 17)

🤖 Agentic Function Calling

final weatherTool = AITool(
  name: 'get_weather',
  description: 'Get weather for a location',
  parameters: {
    'type': 'object',
    'properties': {
      'location': {'type': 'string', 'description': 'City name'}
    },
    'required': ['location']
  },
  execute: (args) async {
    return 'Sunny, 25°C in ${args['location']}';
  },
);

// ConversationManager will automatically:
//  1. Send the tool definition to the AI
//  2. Parse the AI's function call request
//  3. Execute your Dart callback
//  4. Send the result back to the AI
//  5. Return the final summarized answer
final response = await manager.send(convId, 'What is the weather in Cairo?',
  tools: [weatherTool],
);

🎨 UI Widget

AIChatWidget(
  provider: myProvider,
  systemPrompt: 'You are a helpful assistant.',
  theme: AIChatTheme.dark(),
  tools: [weatherTool],
)

📄 RAG (Retrieval-Augmented Generation)

final chunker = DocumentChunker(chunkSize: 500, chunkOverlap: 100);
final docs = chunker.splitText(longDocument);

final vectorStore = InMemoryVectorStore(embedder: openai);
await vectorStore.addDocuments(docs);

final results = await vectorStore.similaritySearch('flutter widgets', limit: 3);

🎤 Audio

// Speech-to-Text (Whisper)
final text = await openai.speechToText(audioBytes, mimeType: 'audio/mp3');

// Text-to-Speech
final audioBytes = await openai.textToSpeech('Hello world', voice: 'nova');

📋 License

MIT License — see LICENSE for details.