flutter_adk

English | 한국어 | 日本語 | 中文

Flutter facade package for ADK Dart core runtime.

What This Package Provides

• Re-exports package:adk_dart/adk_core.dart so Flutter apps can use the Web-safe ADK surface through package:flutter_adk/flutter_adk.dart.
• Exposes core runtime APIs needed for Flutter app usage:
  • Agent / LlmAgent
  • SequentialAgent / ParallelAgent / LoopAgent
  • Runner / InMemoryRunner
  • FunctionTool
  • AgentTool
  • McpToolset (remote MCP over Streamable HTTP)
  • SkillToolset + inline Skill
  • Gemini built-in retrieval tools: UrlContextTool, VertexAiSearchTool, VertexRagRetrievalTool
  • Gemini (BYOK-style key injection)
• Includes a Flutter plugin scaffold registered for all major Flutter platforms:
  • Android
  • iOS
  • Web
  • Linux
  • macOS
  • Windows

When To Use flutter_adk

Use flutter_adk when:

• You are building a Flutter app and want one import that works across mobile, desktop, and web.
• You want the web-safe ADK runtime surface (adk_core) without pulling in VM-only APIs by default.

Use another package when:

• You are building VM/CLI agents, tools, or servers: use adk_dart (or adk for shorter imports).

Design intent:

• flutter_adk is not just a wrapper name; it is the Flutter-oriented compatibility layer.
• It prioritizes consistent multi-platform behavior in Flutter by exposing the web-safe runtime surface (adk_core) instead of the full VM-only API set.

Package Links

• flutter_adk: Flutter-focused package for web-safe ADK usage across Flutter platforms.
• adk_dart: Core runtime package with the full ADK Dart VM/CLI API surface.
• adk: Short-name facade package that re-exports adk_dart.

ADK 2.0 Compatibility

This package is fully aligned with ADK 2.0, supporting v2 Workflows (declarative node-graph scheduling, conditional routing, and state merging) and GCP Managed Agents (including ManagedAgent and RemoteMcpServer configuration mapping) for hybrid on-device and cloud execution.

Installation

flutter pub add flutter_adk

Or add to your pubspec.yaml:

dependencies:
  flutter_adk: ^2026.8.17+3

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)

You can run AI agents completely offline on local models without any 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),
      ),
    );
  }
}

Built-in UI Components & Controllers

flutter_adk provides 12+ ready-to-use widgets and reactive controllers:

• AdkDevStudioView: Turnkey ADK Web / Dev Studio inspector in Flutter (Tabs for Playground, Live Logger, Agent Graph, and State).
• AdkAgentLoggerView: Real-time agent I/O logger (User inputs, LLM responses, Tool calls, State deltas, Latency, Token counts).
• AdkChatView: Full-featured chat UI with autoscroll, input bar, prompt suggestions, and voice triggers.
• AdkChatController: ChangeNotifier-based state controller for managing messages and streaming turns.
• AdkFloatingChatButton: Floating Action Button (FAB) that opens an AI assistant modal bottom sheet.
• AdkPromptSuggestionsBar: Horizontal scrollable bar with quick-tap prompt chips.
• AdkToolCallCard: Expandable card displaying tool name, execution status, JSON arguments, and results.
• AdkConfirmationBanner & AdkConfirmationDialog: Human-in-the-loop (HITL) approval banner and modal dialog.
• AdkSessionDrawer: Chat session history sidebar with create, switch, and delete actions.
• AdkAgentHierarchyBadge: Breadcrumb badge showing active agent in a multi-agent hierarchy.
• AdkWorkflowProgressIndicator: Step-by-step workflow timeline progress tracker.
• AdkVoiceMicButton & AdkAudioWaveVisualizer: Interactive voice input button and animated waveform visualizer.
• AdkTokenUsageBadge: Compact badge displaying prompt and completion token counts.
• AdkMessageBubble: Material 3 message bubble for user, model, and tool outputs.
• AdkTypingIndicator: Smooth pulsing typing indicator.
• AdkEventStreamBuilder: Reactive widget builder for listening directly to Stream<Event>.

Full Runtime Surface

• For VM/CLI-only APIs, import package:adk_dart/adk_dart.dart.

Platform Scope (Current)

• flutter_adk is a Flutter-focused runtime surface built on top of adk_core.
• It targets single-import usage across Flutter platforms while leaving VM/CLI-only APIs outside this package.

Status legend:

• Y Supported
• Partial Supported with caveats
• N Not supported

Feature	Android	iOS	Web	Linux	macOS	Windows	Notes
flutter_adk single import (package:flutter_adk/flutter_adk.dart)	Y	Y	Y	Y	Y	Y	Re-exports the Web-safe adk_core surface.
Agent runtime (Agent, Runner, workflows)	Y	Y	Y	Y	Y	Y	In-memory orchestration path is cross-platform.
Gemini model usage	Y	Y	Partial	Y	Y	Y	Web requires BYOK/CORS/security policy consideration.
Built-in model tools (UrlContextTool, Vertex retrieval)	Y	Y	Y	Y	Y	Y	Tool execution is handled by Gemini/Vertex backends.
MCP Toolset via Streamable HTTP	Y	Y	Y	Y	Y	Y	Works with remote MCP HTTP servers.
MCP Toolset via stdio (StdioConnectionParams)	Partial	Partial	N	Y	Y	Y	Web cannot spawn local processes; mobile runtime support can depend on sandbox/process policy.
Skills (Skill, SkillToolset) with inline definitions	Y	Y	Y	Y	Y	Y	Inline skills are platform-agnostic.
Directory-based skill loading (loadSkillFromDir)	Y	Y	N	Y	Y	Y	Web throws UnsupportedError for filesystem-based loading.
Plugin channel helper (FlutterAdk().getPlatformVersion())	Y	Y	Y	Y	Y	Y	Uses platform channel / browser user-agent path.
VM/CLI tooling (adk executable, dev server, CLI deploy path)	N	N	N	N	N	N	Not part of the Flutter package surface.

Reference matrix and rollout notes:

• knowledge/2026-03-01_18-20-00_flutter_adk_platform_support_matrix.md

Limitations

• Features requiring dart:io, dart:ffi, or dart:mirrors are outside the current flutter_adk surface.
• MCP stdio transport (StdioConnectionParams) is not supported on Web.
• Directory-based skill loading (loadSkillFromDir) is not supported on Web. Use inline Skill definitions.
• For browser BYOK (user-entered API key), document security risks before production rollout.