✅ Repo verified: Hackathon_7March — pushed from version1.1 branch

Run and deploy your AI Studio app

This contains everything you need to run your app locally.

Run Locally

Prerequisites: Node.js

1. Install dependencies: npm install
2. Set the required API keys in .env:
   • GEMINI_API_KEY: For Google Gemini AI.
   • EXA_API_KEY: For Exa search grounding.
   • GMI_CLOUD_API_KEY: (Optional) for GMI Cloud models.
3. Run the app (Frontend + Backend): npm run dev

The app will be available at http://localhost:3000. The backend runs on localhost:8080 (proxied via /api).

Vercel Deployment

This project is optimized for deployment on Vercel. It uses Vercel Serverless Functions for the backend logic.

Deployment Steps

1. Connect to Vercel: Push your code to GitHub/GitLab/Bitbucket and import it as a New Project on Vercel.
2. Environment Variables: Add the following secrets in the Vercel Dashboard:
   • GEMINI_API_KEY
   • EXA_API_KEY
   • GMI_CLOUD_API_KEY (if using GMI models)
3. Build Settings: Vercel should automatically detect Vite. Use the default settings:
   • Framework Preset: Vite
   • Build Command: npm run build
   • Output Directory: dist
4. Deploy: Vercel will build the frontend and automatically set up the serverless functions in the api/ directory.

Project Structure (Vercel-aligned)

• src/: React frontend (Vite).
• api/: Serverless functions (Express) handling /api/* routes.
• vercel.json: Routing and configuration.
• package.json: Unified dependencies and scripts.

GMI Cloud Integration

Atlas supports GMI Cloud as an alternative LLM provider alongside Google Gemini. GMI Cloud offers an OpenAI-compatible inference API with models such as DeepSeek-R1, Llama 3.3 70B, and Qwen 2.5 72B.

How it works

Browser  →  POST /api/gmi/chat  →  Express backend  →  https://api.gmi-serving.com/v1

The GMI_CLOUD_API_KEY is kept on the backend so it is never exposed in the browser bundle.

Setup

1. Obtain a GMI Cloud API key at https://app.gmi-serving.com/api-keys

2. Add it to your root .env file or Vercel environment variables:

   GMI_CLOUD_API_KEY=your_key_here

3. GMI Cloud logic is now integrated into the api/ serverless functions. Local development automatically starts the server.

4. In the Atlas chat UI, open the model selector and choose any GMI Cloud model:

   • deepseek-ai/DeepSeek-R1
   • meta-llama/Meta-Llama-3.3-70B-Instruct
   • Qwen/Qwen2.5-72B-Instruct

Available API endpoint

Method	Path	Description
POST	/api/gmi/chat	Proxies a chat completion request to GMI Cloud

Request body:

{
  "model": "deepseek-ai/DeepSeek-R1",
  "messages": [
    { "role": "system", "content": "..." },
    { "role": "user",   "content": "Suggest beach destinations under $2000" }
  ]
}

Response:

{ "content": "{ ... atlas JSON response ... }" }