flutter_adk #

English | 한국어 | 日本語 | 中文

Flutter facade package and turnkey UI Kit for the ADK Dart core runtime.

What This Package Provides #

• Single Import Facade: Re-exports package:adk_dart/adk_core.dart with Web-safe APIs for seamless use across iOS, Android, macOS, Windows, Linux, and Flutter Web (WASM / dart2wasm).
• Turnkey UI Kit: Complete suite of AI chat, developer studio, logger, tool inspectors, workflow progress bars, and voice interaction widgets.
• Pluggable Persistence: 2-line storage adapters (AdkStorageSessionService) for SharedPreferences, FlutterSecureStorage, SQLite, and Hive.
• Core Runtime:
  • Agent / LlmAgent (Default: gemini-3.7-flash, local LiteLlm Ollama/LM Studio)
  • Workflows: SequentialAgent, ParallelAgent, LoopAgent, ManagedAgent
  • Tools: FunctionTool, AgentTool, McpToolset (Streamable HTTP), SkillToolset
  • Retrieval: UrlContextTool, VertexAiSearchTool, VertexRagRetrievalTool

Installation #

flutter pub add flutter_adk

Or add to your pubspec.yaml:

dependencies:
  flutter_adk: ^2026.8.17+7

Quick Start (Turnkey Chat UI) #

Drop an interactive AI agent chat screen into your Flutter app with just a few lines of code:

import 'package:flutter/material.dart';
import 'package:flutter_adk/flutter_adk.dart';

void main() => runApp(const MyApp());

class MyApp extends StatelessWidget {
  const MyApp({super.key});

  @override
  Widget build(BuildContext context) {
    // 1. Define your AI Agent (Default model is gemini-3.7-flash)
    final agent = LlmAgent(
      name: 'assistant',
      model: 'gemini-3.7-flash',
      instruction: 'You are a helpful and friendly Flutter assistant.',
    );

    return MaterialApp(
      home: Scaffold(
        appBar: AppBar(title: const Text('ADK Flutter Chat')),
        // 2. Embed the turnkey chat view or use AdkDevStudioView(agent: agent)
        body: AdkChatView(
          agent: agent,
          inputPlaceholder: 'Ask anything...',
        ),
      ),
    );
  }
}

Local LLM Quickstart (Ollama / LM Studio) #

Run AI agents completely offline on local models without cloud API keys:

import 'package:flutter/material.dart';
import 'package:flutter_adk/flutter_adk.dart';

void main() => runApp(const LocalAiApp());

class LocalAiApp extends StatelessWidget {
  const LocalAiApp({super.key});

  @override
  Widget build(BuildContext context) {
    // Connect to local Ollama (http://localhost:11434/v1) or LM Studio (http://localhost:1234/v1)
    final localModel = LiteLlm(
      model: 'ollama_chat/gemma2:2b',
      baseUrl: 'http://localhost:11434/v1',
    );

    final agent = Agent(
      name: 'local_assistant',
      model: localModel,
      instruction: 'You are a fast, private offline AI assistant.',
    );

    return MaterialApp(
      home: Scaffold(
        appBar: AppBar(title: const Text('Offline Local LLM Assistant')),
        body: AdkChatView(agent: agent),
      ),
    );
  }
}

📚 Widget & API Specification #

1. Turnkey Views & Developer Tools #

2. State Controller & Storage Services #

AdkChatController

ChangeNotifier-based controller managing conversation history, streaming responses, tool executions, and storage synchronization.

AdkWorkflowController

Reactive controller for executing, pausing (HITL approval), resuming, and tracking step progress in multi-agent workflows and pipelines.

final wfCtrl = AdkWorkflowController(workflowAgent: mySequentialPipeline, initialSteps: steps);
await wfCtrl.execute();
wfCtrl.approveAndResume(stepId: 'step_2', input: {'approved': true});

AdkVoiceController

Reactive controller for managing real-time microphone recording, audio levels (decibels), speech transcripts, and speaking states.

final voiceCtrl = AdkVoiceController();
await voiceCtrl.startListening();
voiceCtrl.toggleMute();
await voiceCtrl.stopListening();

AdkSessionController

Dedicated multi-session controller for loading, creating, switching, renaming, deleting, and searching chat sessions with Key-Value storage sync.

final sessionCtrl = AdkSessionController(storage: storage);
await sessionCtrl.loadAllSessions();
await sessionCtrl.createNewSession(title: 'New Chat');
sessionCtrl.switchSession('session_123');

AdkSmartFormController

Reactive controller for conversational form filling, field auto-extraction from tool arguments, required field validation, and submission.

final formCtrl = AdkSmartFormController(initialFields: fields, onSubmitted: (data) => ...);
formCtrl.populateFromMap({'user_name': 'Alice', 'date': '2026-08-20'});
await formCtrl.submit();

AdkAgentLoggerController

Telemetry and I/O logging controller for buffering agent events, filtering by category/search, and exporting logs to formatted JSON.

final loggerCtrl = AdkAgentLoggerController(maxLogEntries: 500);
loggerCtrl.setCategory(AdkLogCategory.toolCall);
final String jsonLogs = loggerCtrl.exportJson();

AdkAgentManagerController (Central Agent Fleet Manager)

Central registry controller for registering, toggling (ON/OFF), hot-swapping prompts, and tracking cumulative token and latency telemetry across all agents.

final managerCtrl = AdkAgentManagerController();
managerCtrl.registerAgent(researchAgent, id: 'researcher', tags: ['research', 'web']);
managerCtrl.toggleAgent('researcher', true);
print('Avg Latency: ${managerCtrl.getAgent("researcher")?.metrics.avgLatencyMs}');

AdkStorageSessionService

Plug-and-play session persistence backed by any key-value store (e.g. shared_preferences, flutter_secure_storage).

// 2-line SharedPreferences integration:
final sessionService = AdkStorageSessionService.custom(
  read: (key) => prefs.getString(key),
  write: (key, value) => prefs.setString(key, value),
  delete: (key) => prefs.remove(key),
  getKeys: ({prefix = ''}) => prefs.getKeys().where((k) => k.startsWith(prefix)).toList(),
);

AdkTheme & AdkChatThemeData (Theming System)

Wrap your app or screen with AdkTheme to customize colors, corner radiuses, paddings, and font styles across all widgets:

AdkTheme(
  data: const AdkChatThemeData(
    userBubbleColor: Colors.deepPurple,
    modelBubbleColor: Color(0xFFF1F5F9),
    userTextColor: Colors.white,
    modelTextColor: Colors.black87,
    borderRadius: BorderRadius.all(Radius.circular(20.0)),
    inputBackgroundColor: Colors.white,
    sendButtonColor: Colors.deepPurple,
    showTimestamp: true,
    showAvatars: true,
  ),
  child: AdkChatView(agent: myAgent),
)

3. Interactive UI Components & Builder Slots #

4. UI Data Models (package:flutter_adk/flutter_adk.dart) #

• AdkChatMessage: Immutable chat message (isUser, isModel, isTool, isSystem, isStreaming, thought, toolArgs, toolResult, metadata).
• AdkToolCallInfo: Tool execution state model (callId, toolName, arguments, result, status, durationMs).
• AdkWorkflowStep: Workflow step model (id, label, description, status, durationMs, output).
• AdkSessionInfo: Session history model (id, title, messageCount, lastMessagePreview, updatedAt).
• AdkPromptSuggestion: Suggestion model (text, label, icon, category).
• AdkTokenUsage: Token telemetry model (promptTokens, completionTokens, totalTokens, formattedTotal).
• AdkVoiceState: Real-time voice state model (status, decibels, isMuted, isActive).

Platform Support Matrix #

License #

Apache License 2.0. Developed with ❤️ by Deepmind & ADK Labs.