Your unsorted photo pile, now a RAG knowledge base

What it does

Album AI ingests images from a local folder and automatically generates metadata using gpt-4o-mini and Anthropic’s Haiku. It vectorizes that text with OpenAI’s Embedding API, stores everything in PostgreSQL, and exposes both a traditional search endpoint and a RAG chat interface so you can query the collection in plain language. The authors also pitch it as an image knowledge base to feed downstream LLM content generation.

The interesting bit

The whole thing was built in under 24 hours by a photography enthusiast tired of manually cataloging terabytes of shots. It is essentially glue code—vision APIs generate captions, embedding APIs handle vectors, Postgres stores the results, and a thin layer provides search and chat—but the assembly is the point. It treats your local image folder as a living document corpus rather than a static gallery.

Key highlights

• Uses gpt-4o-mini and Haiku for automatic image metadata extraction.
• Vectorizes metadata via OpenAI’s Embedding API for semantic search and RAG retrieval.
• Exposes two APIs: a traditional search endpoint and a conversational chat endpoint.
• Supports Docker-based deployment to platforms like Render.
• Licensed under Apache 2.0, though the authors ask you to contact them for commercial use.

Caveats

• Server deployment docs are still a work in progress; the authors recommend local runs for now.
• Requires live OpenAI and Anthropic API keys, and possibly a proxy, so it does not run fully offline.
• Commercial use requires explicit contact with the authors despite the Apache 2.0 license.

Verdict

Worth a spin for photography enthusiasts with unwieldy local archives and developers who want a reference RAG pipeline over images. Skip it if you need a fully offline solution or a polished, production-grade photo manager.