Architecture
Last updated
Was this helpful?
Last updated
Was this helpful?
Arize is a highly scalable and flexible platform built to meet the demands of both small teams and large enterprises.
Portable - Built with a Kubernetes-first approach and deployable across all major cloud providers and private cloud environments.
Resilient - Ensures data durability through persistent cloud storage, with cloud-based replication across multiple regions.
Scalable - Designed for growth, allowing teams to scale individual components independently and adapt to evolving organizational needs.
The core components of Arize are organized into the following categories:
Front End & APIs - The Arize Web Client serves as the main interface for users, offering UI access, user management, dashboards, and all workflows critical to ML and LLM observability. It is backed by a suite of LLM and ML analytics services exposed through REST and GraphQL interfaces.
Backend Services - Arize supports flexible ingestion methods, accepting data in various formats (e.g., files, tables, OTEL traces) through both push and pull mechanisms, allowing seamless integration with diverse data pipelines and external clients.
Database - At the heart of Arize is ArizeDB, an advanced OLAP database specifically optimized for ML and LLM workloads. Arize also utilizes a local or cloud-hosted PostgreSQL database for operational and metadata storage.
Storage - All observability and operational data is durably persisted to cloud storage to ensure high availability and resilience.
Arize seamlessly integrates with virtually any Kubernetes environment, whether in public or private clouds. It offers out-of-the-box support for popular cloud storage solutions, including AWS S3, GCP GCS, and Azure Blob.
For private cloud deployments without access to traditional cloud storage providers, Arize supports MinIO, Ceph and other S3 compatible solutions. Additionally, it is fully compatible with advanced networking solutions such as Istio and Cilium.
Arize On-Prem ensures that all sensitive data remains secure within the customer’s Virtual Private Cloud (VPC), with strict controls to prevent any confidential information from leaving the security perimeter.
To provide the best possible experience, Arize is typically deployed in a managed configuration, allowing Arize engineers to handle the installation, maintenance, and ongoing support throughout the application's lifecycle. This deployment model includes three key components:
Application Monitoring via Arize Call Home
Image Repository at ch.hub.arize.com or local Artifactory, JFrog, ECR, etc
Kubectl Access for managing the cluster
Custom deployment configurations are available upon request.
