ShitPostr

Meme search and management app with semantic search. Built for homelab.

Inspired by meme-search.

Screenshots

Grid	Detail	Editor

Quick Start

bun install && bun run dev     # Development
docker compose up              # Production

Features

• Semantic Search: Vector embeddings (all-MiniLM-L6-v2) + filename matching
• AI Descriptions: Ollama vision model describes memes automatically
• Meme Editor: Add text overlays, choose fonts/colors
• Zipline Sharing: One-click upload edited memes to Zipline
• Multi-Directory: Mount multiple meme collections
• Mobile Friendly: Responsive layout for phone/tablet

Ollama Integration

Generates searchable descriptions for memes using vision AI.

1. Run Ollama with a vision model: ollama pull llava:7b
2. Set OLLAMA_URL (default: http://localhost:11434)
3. Click "Generate" on a meme or use batch generate in Settings

The model analyzes each image and creates a text description used for semantic search.

Zipline Integration

Share edited memes to your Zipline instance.

1. Set ZIPLINE_URL to your Zipline upload endpoint
2. Set ZIPLINE_TOKEN to your API token
3. Edit a meme → click "Share to Zipline" → URL copied to clipboard

Uploads

Drag & drop or click to upload images via the UI. Files save to UPLOAD_DIR (default /data/memes/uploads) and are automatically added to the database with "pending" status.

Scanning

Scan directories to import existing meme collections:

curl -X POST http://localhost:3000/api/scan \
  -H 'Content-Type: application/json' \
  -d '{"directory": "/data/memes"}'

The scanner recursively finds images (jpg, png, gif, webp, avif), adds them to the database, and extracts titles from filenames. Use Settings UI to trigger scans or batch-generate AI descriptions.

Architecture

File	Purpose
index.ts	All backend: Hono server, API routes, DB, embeddings, scanner
index.html	All frontend: vanilla JS/CSS, grid, modals, meme editor
schema.sql	PostgreSQL + pgvector schema
docker-compose.yml	App + PostgreSQL containers

Stack

• Runtime: Bun
• Server: Hono
• Database: PostgreSQL 17 + pgvector
• Embeddings: Xenova all-MiniLM-L6-v2
• Vision AI: Ollama llava:7b (optional)
• Frontend: Vanilla JS

Environment Variables

DATABASE_URL=postgres://postgres:postgres@localhost:5432/shitpostr
OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=llava:7b
UPLOAD_DIR=/data/memes/uploads
PORT=3000

# Zipline (optional)
ZIPLINE_URL=https://your-zipline.com
ZIPLINE_TOKEN=your-token

# Multiple source directories (comma-separated)
STATIC_DIRS=/data/memes,/data/imgflip

Multiple Source Directories

Mount multiple meme collections as separate Docker volumes:

services:
  shitpostr:
    environment:
      - STATIC_DIRS=/data/memes,/data/imgflip
    volumes:
      - meme-data:/data/memes
      - imgflip-data:/data/imgflip

Then scan each directory:

curl -X POST http://localhost:3000/api/scan \
  -H 'Content-Type: application/json' \
  -d '{"directory": "/data/imgflip"}'

Development

Built with Claude Code.

License

MIT