dartantic_llamadart

Local GGUF model support for Dartantic AI using the llamadart engine.

This package allows you to run fully offline AI agents in Flutter and Dart applications by leveraging local GGUF models.

Features

• Local Inference: Run models entirely on-device without an internet connection.
• Dartantic Interface: Seamlessly integrates with the Dartantic AI ecosystem.
• GPU Acceleration: Inherits hardware acceleration support from llamadart.
• Tool Calling: Supports structured tool calls using XML-tag wrapper format (<tool_call>).

Getting Started

1. Add Dependencies

Add dartantic_llamadart to your pubspec.yaml:

dependencies:
  dartantic_ai: ^1.0.0
  dartantic_llamadart: ^0.1.0

2. Download a GGUF Model

Download a GGUF model (e.g., Llama 3.2 1B) from Hugging Face and place it in your project or a local directory.

3. Initialize the Provider

import 'package:dartantic_ai/dartantic_ai.dart';
import 'package:dartantic_llamadart/dartantic_llamadart.dart';

void main() async {
  final provider = LlamadartProvider(
    name: 'llamadart',
    displayName: 'Local Llama',
    modelPath: 'path/to/your/model.gguf',
  );

  final agent = Agent(
    provider: provider,
    instructions: 'You are a helpful assistant running locally.',
  );

  final response = await agent.prompt('Hello! Who are you?');
  print(response);
}

Advanced Configuration

You can customize inference parameters via LlamadartChatOptions:

final model = provider.createChatModel(
  defaultOptions: LlamadartChatOptions(
    temp: 0.7,
    nCtx: 4096,
    nGpuLayers: 35, // Offload layers to GPU
  ),
);

Tool Calling

To use tool calling with local GGUF models, include instructions in your system prompt to use the <tool_call> XML format:

final messages = [
  ChatMessage(
    role: ChatMessageRole.system,
    parts: [TextPart('''You have access to tools. When you need to call a tool, output:
<tool_call>{"tool_name": {"param1": "value1"}}</tool_call>

Example: <tool_call>{"search": {"query": "weather"}}</tool_call>
''')],
  ),
  ChatMessage(role: ChatMessageRole.user, parts: [TextPart('Search for AI news')]),
];

The model will output tool calls wrapped in <tool_call> tags, which are automatically parsed into ToolPart objects.

License

BSD-3-Clause - See LICENSE for details.