Open-Fiesta

An open-source, multi-model AI chat playground built with Next.js App Router. Switch between providers and models, compare outputs side-by-side, and use optional web search and image attachments.

Features

• Multiple providers: Gemini, OpenRouter (DeepSeek R1, Llama 3.3, Qwen, Mistral, Moonshot, Reka, Sarvam, etc.)
• Selectable model catalog: choose up to 5 models to run
• Web search toggle per message
• Image attachment support (Gemini)
• Clean UI: keyboard submit, streaming-friendly API normalization

Tech Stack

• Next.js 14 (App Router, TypeScript)
• Tailwind CSS
• API routes for provider calls

Quick Start

1. Install deps

npm i

2. Configure environment Create .env.local with the keys you plan to use:

# OpenRouter (recommended for most free models)
OPENROUTER_API_KEY=... 

# Gemini (for Gemini models and image input)
GOOGLE_GENERATIVE_AI_API_KEY=...

3. Run dev server

npm run dev
# open http://localhost:3000

Environment Variables

• OPENROUTER_API_KEY: API key from https://openrouter.ai (required for OpenRouter models)
• GOOGLE_GENERATIVE_AI_API_KEY: API key from Google AI Studio (required for Gemini models)

You can also provide an API key at runtime in the UI’s Settings panel.

Project Structure

• app/ – UI and API routes
  • api/openrouter/route.ts – normalizes responses across OpenRouter models; strips reasoning, cleans up DeepSeek R1 to plain text
  • api/gemini/route.ts, api/gemini-pro/route.ts
• components/ – UI components (chat box, model selector, etc.)
• lib/ – model catalog and client helpers

Notes on DeepSeek R1

Open-Fiesta post-processes DeepSeek R1 outputs to remove reasoning tags and convert Markdown to plain text for readability while preserving content.

License

This project is licensed under the MIT License. See LICENSE for details.

Acknowledgements

• Model access via OpenRouter and Google