Positron IDE Now Runs on Amazon SageMaker for Unified Data Science

Data science teams can now run Positron, Posit's integrated development environment, directly on Amazon SageMaker AI, eliminating the need to switch between separate tools for data access, analysis, model training, and deployment.

The integration allows data scientists to work within a single browser-based environment that combines governed AWS data sources, R and Python sessions, model deployment capabilities, and application development tools. Positron runs as a custom image built on Amazon SageMaker Distribution, which platform administrators deploy to their SageMaker Studio domains.

How the integration works

When data scientists launch a Positron Space in SageMaker Studio, the environment inherits the Space execution role's permissions. This architecture means users can query Amazon Athena, access the AWS Glue Data Catalog, and read from Amazon S3 without managing credentials or API keys. Access control follows the role's defined permissions, and all activity remains within the organization's AWS account.

The system supports parallel workflows through multiple Spaces for independent projects, as well as shared Spaces for team collaboration. Teams can reserve compute capacity using SageMaker AI training plans to ensure availability for scheduled workloads.

Posit Assistant, the company's AI coding helper, can use Amazon Bedrock as its model provider when configured. This keeps AI assistance running on models within the customer's own AWS account and region, with content encrypted and not shared with external model providers.

Demonstrated workflow

AWS and Posit documented a complete workflow using a synthetic 50,000-loan portfolio. The demonstration started with Posit Assistant identifying a table in the AWS Glue Data Catalog and generating an Athena query that returned sample rows in under one second while scanning 2.18 MiB of data.

The workflow then moved data into R for validation and feature engineering, creating debt-to-income and log-income variables. After excluding 1,500 incomplete records, the team trained an XGBoost classifier in Python on 40,000 rows with three features. The model achieved an AUC of 0.834 on held-out data, with a 12.3 percent observed default rate in the highest-risk decile.

The same project deployed a real-time SageMaker AI endpoint, built a Shiny for Python application to invoke it, and generated a Quarto report documenting the complete analysis—all without leaving the Positron environment.

During the Assistant session, the system recorded 6,657,942 tokens including 6,118,411 cache-read tokens, demonstrating 92.5 percent cache efficiency. These figures reflect one session and should not be treated as general cost benchmarks.

Deployment requirements

Organizations need a Posit license grant and administrator permissions to build the Positron image, push it to Amazon Elastic Container Registry, and attach it to their SageMaker Studio domain. The Space execution role must have appropriate permissions for Athena and the Glue Data Catalog. Amazon Bedrock model access is required only when using Posit Assistant.

The demonstration used an ml.t3.xlarge instance. Production deployments require validation of network design, least-privilege permissions, logging, and operational ownership alongside the Posit preview terms.

Why it matters

Data science workflows typically fragment across multiple tools, each requiring separate authentication, environment setup, and context switching. By consolidating governed data access, cross-language analysis, managed deployment, and application development in one authenticated environment, the integration reduces friction in moving from exploration to production. The architecture keeps credentials, data, and AI assistance within the customer's AWS perimeter, addressing common security and compliance requirements for regulated industries.

Details were first reported by AWS in a Machine Learning blog post authored by Abhishek Maligehalli Shivalingaiah, Arantza Rodriguez, Arkaprava De, Sam McIntyre, and Sri Aakash Mandavilli.