DeepFish is a visual AI workflow composer platform that democratizes access to powerful generative AI models. It allows users to combine models from providers like Fal.ai and Replicate into sophisticated pipelines through a node-based interface—think "Zapier for AI" or a web-based ComfyUI.

• Visual Composer: Drag-and-drop node editor to build complex AI pipelines.
• Multi-Modal: Support for Text, Image, Audio, Video, and 3D generation.
• Model Integrations: First-class support for Fal.ai and Replicate models.
• Workflow Library: Import, share, and fork community workflows.
• Credit System: Built-in credit management and Stripe integration for billing.

• Framework: Next.js 15 (App Router)
• Language: TypeScript
• State: Zustand + React Query
• API: tRPC
• Database: PostgreSQL + Drizzle ORM
• Auth: Clerk
• Queues: Inngest (Background processing)
• UI: Tailwind CSS + ShadCN UI + React Flow

• Node.js 18+
• Docker (for local database)
• Accounts for: Clerk, Stripe, Fal.ai, Replicate, Vercel Blob

1. Clone the repo

   git clone https://github.com/noahgsolomon/deepfish
   cd deepfish

2. Install dependencies

   npm install

3. Environment Setup Rename env.example to .env and fill in your keys:

   cp env.example .env

   Note: You will need your own API keys for all services.

4. Database Setup Start the local Postgres container and run migrations:

   # Start the database
   docker compose up -d
   
   # Apply migrations
   npx drizzle-kit migrate

   Note: If you make schema changes, create a new migration with: npx drizzle-kit generate --name="migration_name"

5. Run Development Server

   npm run dev

The core of DeepFish is the Composer (src/app/(app)/(dynamic)/composer). It uses a custom state management layer over React Flow to handle node connections, validation, and execution.

• Current State: Client-directed graph execution.
• Future Goal: A fully persistent, server-side "network" model (akin to TouchDesigner) that supports arbitrary compute nodes and continuous async processing, also allowing for sub-flows (flows/networks which are composed of primitive nodes and also nodes which themselves are flows at infinite recursive depth) allowing workflows to run autonomously even when users are offline.

We use Inngest (src/inngest) to handle workflow executions. This ensures that long-running generation tasks (like video generation) don't block the UI and can handle failures/retries gracefully.

MIT

// also note a lot of the things in public/ folder are outdated and can be removed