Anyscale for developers

This page provides an overview of the Anyscale platform for developers.

Who are Anyscale developers?

Anyscale developers span a number of job titles, but tend to be highly technical users working in the AI and ML space. Many Anyscale users span multiple roles, and the Ansycale platform can help amplify the impact of individuals by simplifying tasks related to infrastructure deployment, monitoring, and production ops.

The Anyscale platform builds on top of Ray, with additional tooling for monitoring, development, and deployment. Developers interact with Python, the CLI, containerfiles, YAML and JSON configuration files, dashboards, and various Anyscale console user interfaces to complete their core job responsibilities.

The following table summarizes the primary tasks for different Anyscale users:

Roles	Primary tasks
Data scientists AI research scientists Reinforcement learning researchers	Build models and run experiments.
ML engineers Software developers Data engineers	Productionalize models, applications, and data processing pipelines.
MLOps DevOps Systems integrators Cloud admins	Configure integrated systems and services, CI/CD, manage infrastructure, keep production applications running.

Depending on the size of your organization and job responsibilities, you might also be responsible for deploying your Anyscale cloud, managing cloud resources or Kubernetes clusters, security, and identity and access management. Anyscale organization owners complete many of these tasks, but any user might have some elevated permissions, especially for resources they create. See Anyscale for admins.

Where should you get started as a developer?

The Anyscale platform provides extensible tooling that you can adapt to many different use cases. Anyscale has designed many products and features with specific users and use cases in mind.

note

If you're the first user of Anyscale in your organization, you might be running in the serverless Anyscale cloud (also called an Anyscale-hosted cloud) deployed during account creation. To access data and integrated systems in your cloud provider account, you should work with a cloud admin to configure and deploy a self-hosted Anyscale cloud on AWS, Google Cloud, or Kubernetes. See Introduction to Anyscale clouds.

The following table helps you navigate to product documentation aligned to common developer tasks and topics:

Topic	Product docs
Development	Use Anyscale workspaces for interactive development on Ray clusters. Anyscale includes dependency management features to simplify handling Python depdencies, system packages, and environment variables. You can deploy development jobs and services directly from a workspace to test code and configurations. For more on development best practices, see Develop Anyscale applications.
Model serving	Use Anyscale services to deploy ML models and other Python code to serving endpoint powered by Ray Serve. Deploy multiple applications to a shared service to maximize hardware utilization.
Schedule and trigger production workloads	Use Anyscale jobs to submit workloads such as batch inference, offline training, and data processing pipelines. Set a job schedule to run your production workloads on a set cadence. Specify a job queue to reuse cloud compute infrastructure for multiple jobs.
CI/CD	Work with your organization admin to configure service accounts to manage integrations with you CI/CD tooling of choice. Configure and use the CLI or SDK with your CI/CD tools. For more details and examples patterns, see CI/CD with Anyscale Jobs and Services.
Observability, monitoring, and debugging	Use workload dashboards to monitor and debug Ray applications. Add custom logging to your applications for additional structured insight into your workloads. View logs and metrics for Anyscale clusters.
Cluster configuration	During development, testing, and production, you manage your compute infrastructure and environment by defining Ray clusters. Container images specify the compute environment, including system packages and Python dependencies. Compute configs control the size and shape of your cluster, including scaling behavior and advanced Ray and cloud infrastructure configurations.