Open-source data engine for processing multimodal AI training data at any scale in a single pipeline.
Daft is an open-source (Apache 2.0) data engine built for AI workloads, enabling teams to process video, images, audio, and sensor data alongside structured metadata in a single dataframe. It runs GPU inference and embeddings alongside CPU decoding in one pipeline, using 5x less memory than alternatives like Spark or Pandas. Built in Rust for speed, Daft supports familiar DataFrame operations (filter, transform, aggregate, write) and is in production at Amazon and Essential AI, with 5,000+ GitHub stars. Daft is a product of Eventual.