Scale applications without rewriting your stack.

Ray gives teams a unified way to parallelize Python, train models, tune experiments, serve AI applications, and run workloads across clusters.

Start Building Install Ray

Quick Start

Install Ray, start a local runtime, and turn ordinary Python functions into distributed tasks.

Install Ray

Use the default extra for the dashboard, cluster launcher, and common runtime dependencies.

pip install -U "ray[default]"

See the full installation guide for Docker, nightly builds, and platform notes.

Start the runtime

Import Ray and start a local runtime. The same application can later connect to a cluster with ray.init(address="auto").

import ray

ray.init()

Run your first task

Add @ray.remote to a Python function, call it with .remote(), and fetch the parallel results with ray.get.

@ray.remote
def square(x: int) -> int:
    return x * x

futures = [square.remote(i) for i in range(8)]
print(ray.get(futures))

Example output

[0, 1, 4, 9, 16, 25, 36, 49]

Choose your next path

Continue with the getting started guide, explore Ray Core, or jump into a higher-level AI library.

Build With Ray

Pick the library that matches your workload, or use Ray Core directly for custom distributed systems.

Ray Core

Use tasks, actors, and the distributed object store to build parallel Python applications.

Ray Data

Load, transform, and run batch inference across large datasets and model pipelines.

Ray Train

Scale PyTorch, Lightning, TensorFlow, and Transformers training jobs across GPUs and nodes.

Ray Tune

Run distributed hyperparameter search with schedulers, search algorithms, and checkpoints.

Ray Serve

Deploy online inference services with autoscaling HTTP and gRPC endpoints.

Ray Clusters

Launch and operate Ray on Kubernetes, VMs, cloud providers, and on-premises hardware.

Production Paths

Move from a first script to production workloads with guides for scaling, serving, and observability.

Deploy model services

Learn the Ray Serve patterns for production deployments, traffic management, and operational tuning.

Run batch inference

Use Ray Data to distribute inference over large datasets with CPUs, GPUs, and streaming execution.

Scale on Kubernetes

Run Ray clusters, jobs, and services on Kubernetes with autoscaling and observability.

Observe workloads

Monitor jobs with the Ray Dashboard, metrics, logs, profiling, tracing, and the State API.

Resources

Learn from examples, join the community, or contribute to the Ray project.

Examples

See end-to-end workflows for data processing, training, tuning, serving, and reinforcement learning.

Community

Find Ray Slack, the discussion forum, meetups, and other ways to connect with users and maintainers.

Contributing

Learn how to file issues, send pull requests, and participate in the Ray open-source project.

Ready to scale your first workload?

Start locally with a few lines of Python, then use the same code on a Ray cluster when your workload grows.

Go to Quickstart

Documentation Index