Self-hosted GLSL shader platform
c9967a17a0
Root cause: Chromium limits ~16 simultaneous WebGL contexts. When scrolling
through a feed of 20+ shader cards, older contexts get silently evicted.
Once a context is lost on a canvas element, getContext('webgl2') returns null
on that same element forever — even after loseContext()/restore cycles.
Solution: The ShaderCanvas component now renders a container div and creates
canvas elements imperatively. When re-entering viewport:
1. Check if existing GL context is still alive (isContextLost)
2. If alive: just restart the animation loop
3. If dead: remove the old canvas, create a fresh DOM element, get a new
context, recompile, and start rendering
This means scrolling down creates new contexts and scrolling back up
replaces dead canvases with fresh ones. At any given time only ~9 visible
canvases hold active contexts — well within Chrome's limit.
Also: 200px rootMargin on IntersectionObserver pre-compiles shaders
before cards enter viewport for smoother scroll experience.
|
||
|---|---|---|
| .bg-shell | ||
| db | ||
| scripts | ||
| services | ||
| .env.example | ||
| .gitignore | ||
| DECISIONS.md | ||
| docker-compose.dev.yml | ||
| docker-compose.override.yml | ||
| docker-compose.yml | ||
| FRACTAFRAG FULL SPEC & SUMMARY.md | ||
| README.md | ||
🔥 Fractafrag
A self-hosted GLSL shader platform — browse, create, generate, and share real-time GPU visuals.
Fractafrag fuses three experiences:
- TikTok-style adaptive feed of living, animated shaders that learns your taste
- Shadertoy-style code editor for writing, forking, and publishing GLSL shaders
- AI generation layer where you describe what you want and the platform writes the shader
Plus a desire queue / bounty board where users express what they want to see, and human creators or AI agents fulfill those requests.
Quick Start
# 1. Clone and configure
cp .env.example .env
# Edit .env with your secrets
# 2. Launch everything
docker compose up -d
# 3. Open
open http://localhost
Architecture
nginx (reverse proxy)
├── / → React frontend (Vite)
├── /api/* → FastAPI backend
└── /mcp/* → MCP server (AI agent interface)
postgres (pgvector/pgvector:pg16) — primary datastore + vector similarity
redis (redis:7-alpine) — cache, rate limiting, job queue
renderer — headless Chromium shader renderer
worker — Celery job processor (render, embed, AI generate)
Tech Stack
| Layer | Tech |
|---|---|
| Frontend | React 18, Vite, Three.js, TanStack Query, Zustand, Tailwind CSS |
| Backend | Python, FastAPI, SQLAlchemy, Pydantic |
| Database | PostgreSQL 16 + pgvector, Redis 7 |
| Jobs | Celery + Redis |
| Renderer | Node.js + Puppeteer (Headless Chromium) |
| MCP | Python MCP SDK, HTTP+SSE transport |
| Payments | Stripe (subscriptions + Connect) |
| Container | Docker Compose, single-stack |
Milestone Roadmap
| Milestone | Focus | Status |
|---|---|---|
| M0 | Infrastructure + Auth | 🚧 In Progress |
| M1 | Core Shader Loop (editor, submit, feed) | ⏳ |
| M2 | Intelligence Layer (MCP, recommendations) | ⏳ |
| M3 | Desire Economy (bounties, fulfillment) | ⏳ |
| M4 | Monetization (Stripe, subscriptions) | ⏳ |
| M5 | AI Generation (prompt → shader) | ⏳ |
Development
# API direct access (dev mode)
http://localhost:8000/api/docs # Swagger UI
http://localhost:8000/health # Health check
# Services
http://localhost:5173 # Vite dev server
http://localhost:3200 # MCP server
http://localhost:3100 # Renderer
License
Private — see DECISIONS.md for project governance.