Meme Search is an open source meme search engine that developers and hobbyists host themselves. It uses AI to index a collection of meme images by their visual content and embedded text, so a meme can be found by what it shows rather than by file name.
A local vision language model reads each image and writes a text description, with a choice of several models. Descriptions can be generated for selected memes or in bulk for a whole library, and they can be edited by hand to improve results. Search combines vector and keyword matching using Postgres with pgvector. The app supports tags and filters, multiple directory paths with rescanning, drag and drop upload of JPG, PNG and WEBP files with up to 50 files at a time, and a dark mode. Processing is local by default, with an optional OpenAI compatible vision provider.
A read only Search API version 1, a local command line tool and an experimental Chromium popup let people retrieve memes outside the web app. The API is token authenticated, and official support covers loopback access only, so exposing it directly to the public internet is unsupported.
Deployment is a single docker compose command that starts the app, database and AI generator. It is intended for people with large personal meme folders who want private, semantic retrieval. It is free and open source, and no paid plans are listed.
