🧠 mindloop-gemini

MindLoop Gemini — a lightweight, type-safe framework for building AI agents with Google's Gemini models.

Core orchestration for Gemini agentic workflows — handling tools, memory, and structured outputs.

✨ Features

• ⚙️ Gemini Native — built specifically for @google/genai
• 🧩 Tool System — define reusable tools using FunctionDeclaration from GenAI SDK
• 💾 Memory System — simple InMemory or FileMemory conversation storage
• 🧱 Extensible Design — perfect for building your own agent SDKs or frameworks
• 📦 Minimal dependencies

🚀 Installation

npm install mindloop-gemini @google/genai zod
# or
yarn add mindloop-gemini @google/genai zod

🧰 Quick Example

import { Agent, InMemory, MindLoopTool } from "mindloop-gemini";
import { Type } from "@google/genai";

const tools: MindLoopTool[] = [
  {
    handler: async ({ city }) => {
      // Your tool logic here
      return {
        success: true,
        data: `It's sunny in ${city}`,
        error: undefined,
      };
    },
    declaration: {
      name: "getWeather",
      description: "Returns a weather summary for a city",
      parameters: {
        type: Type.OBJECT,
        properties: {
          city: {
            type: Type.STRING,
            description: "The city to get weather for",
          },
        },
        required: ["city"],
      },
    },
  },
];

const agent = new Agent({
  apiKey: process.env.GEMINI_API_KEY!,
  tools,
  memory: new InMemory(),
  model: "gemini-2.5-flash",
  verbose: true,
});

const response = await agent.run("What's the weather in Tokyo?");
console.log(response);

🧩 Memory Options

| Type | Description | Example | | ------------ | --------------------------------------------- | -------------------------------------------- | | InMemory | Keeps messages in memory (cleared on restart) | new InMemory() | | FileMemory | Persists messages to a JSONL file | new FileMemory({ filePath: "chat.jsonl" }) |

📜 License

MIT License © 2025 Tanmay Vaij