gemini_agent_tools

Note: This package was developed for use alongside a University final year project, though it was published to make it easier for me to use as well as as an exercies in publishing.

gemini_agent_tools is a Dart/Flutter package that enables automatic code generation for Gemini-style agent tool declarations to be used in with the REST api, specifically the function declaration format specified here.

It allows you to define tools as annotated Dart classes, and a custom build_runner generator produces the correct API-ready JSON schemas automatically.

This eliminates manual boilerplate, keeps tool definitions type-safe, and ensures your agent functions remain in sync with your code.

---

🚀 Features

• 🔧 Annotate classes to define agent tools

• ⚙️ Build-runner code generator creates *.tools.g.dart output

• 📐 Auto-generates:

  • tool names
  • descriptions
  • JSON schema parameters and their types*
  • property ordering
  • required fields

• 🐍 Converts camelCase → snake_case automatically

* Only String, bool and int types currently supported.

---

📥 Installation

Add the package:

dart pub add gemini_agent_tools

And the dev dependency for build_runner:

dart pub add --dev build_runner

---

🧩 Usage

1. Annotate a class with @Tool

import 'package:gemini_agent_tools/tool_annotation.dart';

@Tool(
  name: "save",
  description: "Stores information to agent memory",
  requiredFields: ["agentNotes"],
)
class SaveTool {
  final String agentNotes;

  SaveTool(this.agentNotes);
}

The annotation fields:

Field	Type	Description
name	String	The tool's function name
description	String	Human-readable description
requiredFields (optional)	List<String>	Marks parameters as required

Parameters are generated from class fields automatically.

---

2. Run the generator

dart run build_runner build

This will produce a file:

save_tool.tools.g.dart

---

3. Generated Output Example

const Map saveToolAsMap = {
  "name": "save",
  "description": "Stores information to agent memory",
  "parameters": {
    "type": "object",
    "properties": {
      "agent_notes": {"type": "string"}
    },
    "propertyOrdering": ["agent_notes"],
    "required": ["agent_notes"]
  }
};

This object is ready to pass directly into a Gemini REST API functionDeclarations list.