An open-source, self-hostable meme search engine that uses AI to index and retrieve memes via semantic search.
Meme Search is an open-source meme search engine and finder designed for self-hosting. It uses AI models to analyze meme images, extract text and semantic meaning, and index them for fast retrieval via a web interface. The tool solves the problem of organizing and finding specific memes in large personal collections.
Developers, meme enthusiasts, and communities who want to privately manage and search large personal meme libraries without relying on cloud services.
Developers choose Meme Search because it offers full local processing for privacy, supports multiple AI models for flexibility, and provides a complete self-hosted stack with Docker for easy deployment.
The open source Meme Search Engine and Finder. Free and built to self-host locally with Python, Ruby, and Docker.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
All image processing, embedding, and search happen on your hardware with no data sent externally, as emphasized in the privacy-first philosophy and local processing features.
Supports multiple vision language models from Florence-2 to Moondream2, allowing users to trade off accuracy and resource usage based on their hardware, with options like INT8 quantization for memory-constrained setups.
Uses PostgreSQL with pgvector for fast keyword and semantic search, enabling quick retrieval in large meme collections, as highlighted in the features for streamlined database transactions.
Includes bulk operations, drag-and-drop uploads, dark mode, and directory rescan, making meme management intuitive and user-friendly, with GIF examples in the README demonstrating these tools.
Requires careful volume mounting and configuration in Docker Compose, with inter-container communication tweaks like extra_hosts for Linux users, which can be error-prone and daunting for those unfamiliar with Docker.
AI models demand significant memory (e.g., Moondream2 uses ~5GB) and computational power, making it unsuitable for low-end hardware without compromises like using smaller or quantized models.
Only supports JPG, PNG, and WEBP images, excluding common meme formats like GIFs or videos, which restricts its utility for broader multimedia collections without custom modifications.
Docker images are built manually with no automated CI, and E2E tests must be run locally by contributors, adding maintenance overhead and potential delays for updates and contributions.