---
id: daft
version: "0.7.23"
license: unclear
license_treatment: unclear
maintenance: active
---

# daft — Distributed Dataframes for Multimodal Data

License: unclear · Maintenance: active · Downloads: 1.1M/mo

## What it is and what it does

Daft is a Python-native, Rust-powered distributed dataframe engine designed for AI and multimodal workloads. It processes images, audio, video, and structured data in a single framework, with built-in operations for running LLM prompts, generating embeddings, and classifying data at scale. The engine supports starting locally and scaling to distributed clusters via Ray or Kubernetes, and can access data from S3, GCS, Iceberg, Delta Lake, Hugging Face, and Unity Catalog.

The package depends on pyarrow, fsspec, tqdm, typing-extensions, and packaging. It targets developers working with mixed data types (text, images, audio, video) who need to process them together at scale without JVM overhead. The framework handles memory management and provides sensible defaults to reduce configuration burden.

Use it for:

- Process e-commerce product images alongside structured metadata and run AI inference on them at scale
- Load multimodal datasets from Hugging Face and apply LLM prompts or embeddings to both text and images
- Scale local data processing workflows to distributed Ray clusters without rewriting code
- Combine audio, video, and structured data in a single pipeline for media analysis tasks
- Access and transform data from data lakes (Iceberg, Delta Lake) with native multimodal support

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Daft is a distributed dataframe engine for processing images, audio, video, and structured data at scale, with built-in AI operations and support for multimodal workloads.

Yes, if you need to process multimodal data (images, audio, video, text) at scale with built-in AI operations. The active maintenance, recent releases, and lack of known vulnerabilities are positive signals. Verify the Apache 2.0 license terms match your use case, and confirm that medium install friction (compiled Rust wheels) is acceptable for your deployment environment. Not necessary if you work only with structured data or single-modality inputs.

## Install

pip install daft
uv add daft
poetry add daft

## Installing daft

Before you install: Medium install friction due to compiled Rust components (wheels provided for common platforms). Active maintenance with a recent release (9 days ago) and 5709 repository stars. Requires Python 3.10 or higher.

License in practice: License treatment is unclear in the fact sheet; the description states Apache 2.0, but SPDX and raw license fields are not populated. Verify the actual license terms before adopting in proprietary projects.

Quickstart:

pip install daft
import daft
df = daft.from_pydict({"col": [1, 2, 3]})
df.show()

Requires Python 3.10 or higher; compiled wheels available for macOS (x86_64, arm64), Linux (x86_64, aarch64), and Windows (x86_64).

Verify before relying:

- Whether Apache 2.0 license is correctly applied given unclear license_treatment in metadata
- Performance benchmarks and scaling characteristics on distributed clusters
- Maturity and stability guarantees for production AI workloads

## Package facts

- License: not declared (unclear)
- Python support: supports_current
- Install friction: medium
- Maintenance: active
- Downloads: 1.1M/month (top 5,000 on PyPI)
- Known vulnerabilities: none known

## Tags

distributed dataframe engine, multimodal data processing, image audio video processing, AI inference at scale, structured and unstructured data, Ray Kubernetes distributed computing, Python data engine, multimodal-data, distributed-computing, data-engine

[View on SkillFed](https://skillfed.io/packages/daft) · [View on PyPI](https://pypi.org/project/daft/)