Positron arrives on Amazon SageMaker AI in public preview

Posit has made its Positron IDE available in Amazon SageMaker AI Studio Spaces through a customer-managed custom image. Posit calls the integration a public preview and says it supports SageMaker Studio, with SageMaker Unified Studio support still to come.

For SageMaker users, the release brings R and Python development, AWS data access, model deployment, applications and reporting into one browser-based environment. Administrators must configure the image before it appears as a selectable option in Studio.

The setup follows SageMaker’s custom-image lifecycle. An administrator must place the Positron image in a private Amazon Elastic Container Registry repository in the Studio domain’s AWS Region, register it as a SageMaker image and version with an app image configuration, and attach it to the domain. Users can then select Positron when they create a JupyterLab Space.

The two companies describe the image step differently. AWS says administrators build Posit’s published image definition and push the result to their registry. Posit says administrators pull a published image and copy it into their registry. The official materials do not explain whether the difference depends on the image version or distribution route.

Posit says the commercial offering requires a Posit Workbench Advanced license and checks entitlement through AWS License Manager. AWS’s walkthrough also lists administrator permissions for ECR and SageMaker custom images, a Space execution role that can reach the required AWS services, and an ml.t3.xlarge instance or larger for the demonstrated environment.

Posit’s published 2026.09 Containerfile is based on the CPU version of SageMaker Distribution 4.4.1. It adds Positron Server, R and Python runtimes, Shiny, uv, Quarto, professional database drivers, and a Posit Assistant configuration backed by Amazon Bedrock.

Positron uses the Space execution role. Access to Athena, the Glue Data Catalog, S3 and SageMaker endpoints therefore follows that role’s permissions rather than stored access keys. Bedrock is optional in AWS’s deployment description and is involved when selected as the assistant’s model provider.

AWS demonstrated the integration with a synthetic portfolio of 50,000 loans. In the recorded run, Athena queried the data, R validated features, Python trained an XGBoost classifier, SageMaker hosted a real-time endpoint, a Shiny for Python application invoked it, and Quarto generated a report from the same project.

Those results apply only to the demonstration. AWS said one held-out split produced an AUC of 0.834 and a 12.3% observed default rate in the highest-risk decile. It also reported 6,657,942 tokens, 92.5% cache efficiency and an estimated cost of $6.319 for the recorded Posit Assistant session. AWS said the figure was an assistant estimate, not an AWS invoice or a general cost benchmark.

AWS said the synthetic data and applicant payloads do not represent a production lending system. The demonstration did not establish fairness, calibration, lending suitability, production latency, load behavior, monitoring or regulatory compliance. Its screenshots and metrics should not be treated as general performance or cost benchmarks.