flutter_local_agent_kit 1.0.1

Offline-first AI framework for Flutter. Resident LLM inference, local RAG, and autonomous agents with a Material 3 Chat UI.

Flutter Local Agent Kit 🤖✨

The ultimate offline-first AI framework for Flutter. Build autonomous agents, private RAG systems, and high-performance chat interfaces that run 100% on-device.

---

🌟 Key Features

• 🧠 High-Performance Inference: Native integration with llamadart supporting Llama 3.2, Gemma, and Mistral.
• 🕵️ Autonomous Agents: Built-in ReAct (Reason-Act) loop for tool use (Calculators, APIs, System tasks).
• 📚 Private RAG: local vector database for knowledge injection without cloud dependencies.
• 🎨 Premium UI Components: 120Hz smooth AgentChatView with Markdown, code blocks, and Suggestion Chips.
• 🛡️ Secure & Private: No API keys, no data leaves the device. Perfect for privacy-first enterprise apps.
• ⚙️ Customizable Personas: Easily set custom system prompts for specialized agent behaviors.

---

🚀 Quick Start

1. Initialize the Kit

final kit = FlutterLocalAgentKit();

await kit.initialize(
  modelPath: '/path/to/llama-3.2-1b.gguf',
);

1a. Detect LLM-Only Fallback

await kit.initialize(modelPath: '/path/to/llama-3.2-1b.gguf');

if (!kit.isRagReady) {
  debugPrint('RAG unavailable: ${kit.ragInitializationError}');
}

The kit can still become ready when the optional RAG subsystem fails to boot. Use isRagReady and ragInitializationError to disable knowledge features or surface a degraded-mode message in your app.

2. Run an Autonomous Agent

kit.runAgent("Calculate my tax for 50k salary and tell me the time.").listen((chunk) {
  print(chunk); // Streams "Thought -> Action -> Observation -> Final Answer"
});

3. Use the Premium UI

AgentChatView(
  onMessage: (query) => kit.runAgent(query),
  suggestions: const ['🕵️ Who are you?', '📅 Get Time', '🧮 Solve math'],
  welcomeMessage: "Hello! I am your local AI agent.",
)

4. Inject a Custom Runtime Adapter

final kit = FlutterLocalAgentKit(
  runtimeAdapter: MyKitRuntimeAdapter(),
);

KitRuntimeAdapter lets you swap how LLM and RAG sessions are created. This is useful for testing, custom native integrations, or controlling engine lifecycle outside the default adapter.

---

🛠️ Built-in Tools

• Calculator: High-precision math execution.
• DateTime: Real-time context awareness.
• Custom Tools: Easily extend with BaseTool.

---

📱 Performance (OnePlus 12)

• Model: Llama 3.2 1B (Instruct)
• RAM Usage: ~900MB (Stable)
• Throughput: 45+ tokens/sec
• Latency: <100ms first-token (Native Vulkan/Impeller)

---

📄 License

MIT License. Built with ❤️ for the Flutter Ecosystem (2026).