llm_tool_generator

The code generator for llm_tool: turn any Dart function into an LLM tool with one annotation. No hand-written JSON schemas.

For every @LlmTool() function it generates a ToolDefinition with the JSON Schema for the LLM, argument validation and type-safe dispatch.

The full documentation, including supported types, limits, validation and errors, and how to send tools to OpenAI, Claude, Gemini, MCP or on-device models, is in the llm_tool README.

Quick start

1. Install

dart pub add llm_tool dev:llm_tool_generator dev:build_runner

2. Write a tool

import 'package:llm_tool/llm_tool.dart';

part 'tools.g.dart';

/// Gets the current weather for a city.
@LlmTool()
String getWeather(
  @Param('City name, e.g. Kanpur') String city, {
  @Param('Use Celsius instead of Fahrenheit') bool celsius = true,
}) => 'Sunny, ${celsius ? '31°C' : '88°F'} in $city';

3. Generate

dart run build_runner build

4. Use the generated getWeatherTool

print(getWeatherTool.parametersSchema); // send to your LLM
print(await getWeatherTool({'city': 'Kanpur'})); // Sunny, 31°C in Kanpur

What gets generated

final getWeatherTool = ToolDefinition(
  name: "getWeather",
  description: "Gets the current weather for a city.",
  parametersSchema: {
    "type": "object",
    "properties": {
      "city": {"type": "string", "description": "City name, e.g. Kanpur"},
      "celsius": {
        "type": "boolean",
        "description": "Use Celsius instead of Fahrenheit",
      },
    },
    "required": ["city"],
    "additionalProperties": false,
  },
  requiresConfirmation: false,
  execute: (args) => getWeather(
    args["city"] as String,
    celsius: args["celsius"] as bool? ?? true,
  ),
);

/// Every tool in this file, e.g. to send to an LLM or look up by name.
/// Typed by the tools' common return type, so calling one needs no cast.
final allTools = [getWeatherTool];

The list is named after the file: tools.dart gives allTools, weather_tools.dart gives weatherTools.

For a class marked @LlmToolset(), its @LlmTool methods are generated into an extension, bound to the instance they're called on:

@LlmToolset()
class WeatherTools {
  WeatherTools(this._api);
  final WeatherApi _api;

  /// Gets the current weather for a city.
  @LlmTool()
  Future<String> getWeather(@Param('City name') String city) =>
      _api.current(city);
}

// Generated:
extension WeatherToolsLlmTools on WeatherTools {
  List<ToolDefinition<String>> get llmTools => [
    ToolDefinition(
      name: "getWeather",
      // ...
      execute: (args) => getWeather(args["city"] as String),
    ),
  ];
}

// Use it:
final tools = WeatherTools(api).llmTools;

Supported

• Top-level functions, and instance or static methods of @LlmToolset() classes (generated as an llmTools getter): sync, async (Future<T>) and void.
• String, int, double, num, bool, enum, class and List parameters (lists of any of these, including nested lists): positional or named, nullable or not, with or without defaults.
• Descriptions from @LlmTool(description: ...) or the doc comment.
• Custom JSON names with @Param('...', name: 'game_id'), so your Dart code keeps camelCase while the LLM sees snake case.
• Limits with @Param('...', min: 0, max: 130), minLength, maxLength, pattern, minItems and maxItems. See Limits.
• Class parameters become nested object schemas, built through the class's unnamed constructor. freezed classes work too. See Class parameters.

Anything else is a build-time error with a message explaining the fix: unsupported types, missing descriptions, tool names that LLM providers would reject, generic functions, limits that don't fit the type, and @LlmTool on methods of a class without @LlmToolset().

Troubleshooting

Undefined name 'getWeatherTool': add part 'your_file.g.dart'; and run dart run build_runner build. A @LlmTool method needs @LlmToolset() on its class; without it, in a file with no other top-level annotation, the method is skipped without a message.

The name 'ToolDefinition' is defined in the libraries ...: another package (e.g. flutter_ai_core) also has a ToolDefinition. Import llm_tool with a prefix (as ltc) and use @ltc.LlmTool(); the generated code follows your prefix.

More in the full troubleshooting guide.

Author

Built and maintained by Amit Gupta. Bug reports, ideas and pull requests are welcome on GitHub. If this package saves you time, a like on pub.flutter-io.cn or a star on GitHub helps others find it.