llm_tool_generator 0.8.0

Code generator for llm_tool. Turns @LlmTool-annotated Dart functions into LLM tools for OpenAI, Gemini, Claude and MCP, with JSON schemas and validation.

0.8.0

• Generates toolsets: for each @LlmToolset() class, an extension with an llmTools getter listing its @LlmTool methods (instance and static) as tools bound to that instance, typed by their common return type.
• @LlmTool on a method of a class without @LlmToolset is still a build error, now suggesting @LlmToolset.
• Writes @Param limits into the schema (min → minimum, minLength, pattern, minItems, ...). On a list, value limits apply to its items. A limit that doesn't fit the type, min above max, a negative length or an invalid pattern is a build error.
• Requires llm_tool 0.8.0.

0.7.0

• Renamed from llm_tool_calling_generator (now discontinued), to go with the core package's new name, llm_tool. The builder is now llm_tool_generator:llm_tool, and generated parts are .llm_tool.g.part.
• Recognizes @LlmTool(), and still the deprecated @Tool().
• Generated tools are typed by their function's return type (e.g. ToolDefinition<String>), and the list of all tools by their common return type (e.g. List<ToolDefinition<Command>>), so calling a tool needs no cast.
• Requires llm_tool_calling 0.7.0.

0.6.1

• Added the package homepage (https://amitgp.dev).

0.6.0

• No changes in the generated code. Requires llm_tool_calling 0.6.0, which adds built-in SDK formats and invoke.

0.5.0

• Supports @Param(name: ...): the schema, required and validation use the JSON name, while your function is still called with its Dart parameter names. Invalid or duplicate JSON names are build-time errors.

0.4.0

• The generated code uses your import prefix for ToolDefinition, so import 'package:llm_tool_calling/llm_tool_calling.dart' as ltc; works, and so does another package's ToolDefinition in the same file (e.g. flutter_ai_core).
• Tool names are limited to 63 characters, the firebase_ai limit.

0.3.0

• Generates a list of all tools in each file, named after the file: allTools for tools.dart, weatherTools for weather_tools.dart. If you already declared a variable with that name in the file, rename it.
• README: author section.

0.2.0

• Class parameters become nested object schemas, built through the class's unnamed constructor. Descriptions from @Param or field doc comments; defaults, nested classes, lists of classes, freezed classes and prefixed imports are supported.
• Generated code no longer causes analyzer warnings in your project.
• Tool names must start with a letter or _, and parameter names may only use letters, digits and _, so every schema works with Gemini as well as OpenAI and Claude. Other names are a build-time error.
• List<T> parameters for every supported T, including enums and nested lists. List<int> and List<double> accept any JSON number.
• Enum parameters, sent to the LLM as their value names. Works with nullable enums, defaults, enhanced enums and prefixed imports (import 'units.dart' as u;).

0.1.0

• Initial release.
• Generates a <functionName>Tool ToolDefinition for every @Tool top-level function.
• Parameter types: String, int, double, num, bool, nullable or not, positional or named, with or without defaults.
• Description from @Tool(description:) or the doc comment (/// or /** */).
• Sync, async and void functions.
• Build-time errors for unsupported types, missing descriptions, invalid tool names, generic functions and @Tool on methods.