Pixeltable

Pixeltable is an open-source, unified multimodal data backend built for AI applications. It gives you a Python-native dataframe API to store, query, and version images, video, audio, text, and structured data in one place. Every operation is automatically recorded, so you can replay, fork, and branch your data like code — perfect for RAG pipelines, computer vision datasets, evaluation sets, and agentic workflows. Data is stored in cloud object storage with a Postgres metadata layer, keeping infrastructure simple and durable. Pixeltable handles complex transforms, embeddings, and model inference with incremental computation, so you never recompute what has not changed. It is Apache-2.0 licensed, self-hostable with Docker, and designed for teams that need reproducibility and full control over their AI data stack.