How DataFramer connects to your stack
Two ways to deploy
Both run the same application. They differ in where your data is processed and stored.DataFramer-hosted
Everything runs in DataFramer’s cloud, and your data is stored there, scoped to your company. Model calls run on DataFramer credits or on your own model credential; see AI usage and credits. Nothing to deploy — just sign up.Customer-hosted
Use this when your data cannot leave your own cloud. DataFramer keeps the control plane: the web app, job orchestration, and metadata. You run the data plane in your own cloud account, deployed with our Terraform module: the workers that process your data and the storage they use.
- Your content stays in your account.
- No inbound connections. Every connection that crosses the boundary starts in your account or in your browser.
- Model calls use your own provider credential.
- Same functionality in the web app as the DataFramer-hosted version.
- The API and Python SDK start and track your jobs and return names, statuses, counts, and scores. The Python SDK also uploads and downloads your files and reads and edits your specs, straight to and from your data plane, with the same methods as on the DataFramer-hosted version. It reads generated text the same way: evaluation results, anonymization results, red-team prompts, failure analysis and context. Evaluating a single item with a judge runs in your data plane too: the item is sent to your own gateway, never to DataFramer.