( In Development, pre-alpha ) Something like Figma, but for designing AI systems. Maintained by Shubhamai.

Homepage | Documentation | API Reference

About

AISandbox is a node-based editor that allows creating an architecture of multiple models with inputs and outputs to perform a range of tasks. Think of Figma but for designing AI systems graphically.

Note that is still in development, pre-alpha.

Features

• Graphical Editor - Create a graph of models, with inputs and outputs.
• REST API - Run the graph as an API, with a single click.
• Self Hosted - Entire application can be self-hosted.
• Dashboard - Manage your projects, and view usage statistics.

Documentation

Documentation | API Reference

To run the graph as an API, the code is programmatically generated in the UI, but here's a sample code

from aisandbox import execute_project

project_id: str = "91392237-fc87-48b4-b83c-59c6ce0aad9d"

api_key: str = "ais-4ad62b20-74ab-4f55-a5df-856007db6261"


inputs = [{"id": "TextInputNode-dU-gccjt6igEbAVFnYs4x", "data": {"text": "Hello!"}}]

output = execute_project(project_id, api_key, inputs)

print(output.json())

Tech Stack

Front end

• Nextjs 13 as the web framework.
• shadcn ui for the UI components.
• reactflow for the node editor.
• Lucide for icons.
• zustand for state management.
• TailwindCSS - CSS Framework

Backend

• Supabase for user authentication and serves as the main database.
• Stripe for Payments.
• OpenAI and Replicate to run most of the models.
• Upstash - Rate Limiting, Caching.

Infrastructure

• The site is deployed on cloudflare self hosted on Digitalocean $7 droplet, with cloudflare as reverse proxy.

Self Hosting

(In progress)

To self-host this application ( at least some of it ), follow the steps :

• Fill up the following API keys in .env.example, then rename it to .env.
• Make an account of supabase, and create the following DBs with types provided in
• Run pnpm dev.

Acknowledgements

• Thanks to Resend, Skiff, Supabase and Figma for site UI, design and landing page inspirations.
• Thanks to Sniffnet, nextjs-subscription-payments, tinygrad for readme, code structure.