# Eventual-Inc/Daft

High-performance data engine for AI and multimodal workloads. Process images, audio, video, and structured data at any scale

Repository: https://github.com/Eventual-Inc/Daft
Canonical: https://ross.abutalabs.com/products/daft
Homepage: https://daft.ai
Language: Rust
License: Apache-2.0
License Family: permissive
Topics: machine-learning, python, data-engineering, distributed-computing, rust, big-data, artificial-intelligence, multimodal, arrow, distributed, distributed-systems, embeddings, etl, huggingface, iceberg, parquet, ray, ai-engineering, ai-pipeline
Last push: 2026-08-26T18:42:39+00:00

## Health v2 (maintenance only)
Score: 95/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 86, longevity 100
- inputs: {"age_days": 1591, "days_push": 7, "days_rel": 19, "gap_med": 5, "n_releases_24m": 96}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5730, forks 546 (observed 2026-08-28T04:09:28.303517+00:00)

## What it is
Daft is a high-performance distributed data engine with a Python dataframe API, implemented in Rust, designed for AI and multimodal workloads. It processes images, audio, video, embeddings, and structured data at scale, with built-in model operators and scaling from local to Ray or Kubernetes clusters.

## Use cases
- process images and video at scale for ML training data
- generate embeddings for a vector database
- run LLM extraction over large datasets
- build multimodal AI ETL pipelines
- prepare training-ready datasets from S3 or Iceberg
- distributed dataframe processing without Spark's JVM
- run GPU inference alongside CPU data transforms

## When to choose
- you need Pandas/Spark-like dataframe operations on multimodal data
- you want to mix CPU decoding and GPU inference in one pipeline
- you need to scale from laptop to Ray or Kubernetes clusters
- you work with Parquet, Iceberg, Delta Lake, or Hugging Face datasets

## When to avoid
- you only need small in-memory analytics that Pandas handles fine
- your workloads are purely tabular with no AI or multimodal needs
- you require a JVM-based Spark ecosystem with its integrations

## Facets
- artifact type: library
- maturity: active
- function: etl, data-science, machine-learning, rag, llm-inference, image-processing, audio-processing, video-processing, streaming
- domain: machine-learning, artificial-intelligence, big-data, data-science
- platform: python, rust, cloud, cross-platform
- tags: dataframe, distributed-computing, multimodal, parquet, iceberg, ray, embeddings, gpu, arrow, ai-pipeline, data-engineering, docker, kubernetes

## Member repositories
- Eventual-Inc/Daft (main) score 95

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:28.303517+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:53:23.741212+00:00, confidence not recorded.
  - readme: https://github.com/Eventual-Inc/Daft (fetched 2026-08-28T04:09:28.303517+00:00, sha f79705bb72be)
  - homepage: https://daft.ai (fetched 2026-08-29T08:52:59.628068+00:00, sha 196ff9b9bb82)
- Data as of 2026-08-30T08:39:29.467469+00:00.