Answers before your first prompt.

A playground is a focused visual workspace for trying a model without first building an application. Benchius includes chat, image, video, and document experiences, plus a benchmark for side-by-side comparisons.

For most playgrounds, have these two values ready:

• An inference endpoint for the model deployment or provider you want to call.
• An API key with permission to use that endpoint.

Some recognized HTTPS Microsoft Foundry endpoints also support Microsoft Entra ID authentication. The selected playground will show the authentication modes it can use and any additional model-specific fields.

Open the deployment or model details in Microsoft Foundry or the provider portal. Copy the inference endpoint and one of the credentials shown for that resource. Use the endpoint for model inference, not the portal page URL or an Azure resource-management URL.

Use Microsoft Entra ID when the playground recognizes the Foundry endpoint and your account or application has the required access. Use API key authentication for native provider endpoints, custom endpoints, or deployments that do not support the offered Entra ID flow.

When auto-save is enabled, fields such as ApiKey and ClientSecret are encrypted with ASP.NET Core Data Protection before they are written to the browser's local storage. The stored value is a protected ciphertext, not the API key in readable text. On a later visit, the app uses the matching Data Protection key and playground-specific purpose to decrypt the value and restore the field.

It reduces exposure at rest: casually inspecting local storage, copying the browser profile, or reading a storage backup does not reveal a directly usable API key. The ciphertext is tied to this app's Data Protection keys and purpose, so a different key ring cannot decrypt it.

This is defense in depth, not a replacement for a secret manager. A compromised browser or active app session may still access credentials while they are in use. Rotate exposed keys and follow your organization's credential policy.

Not automatically. Local storage is scoped to the browser profile and site, and the encrypted value also requires the matching Data Protection keys and purpose. If those keys are unavailable or change, the saved credential cannot be restored and you will need to enter it again.

No. Interactive authentication state such as access tokens, refresh tokens, ID tokens, and account cache identifiers is excluded from playground input persistence.

No, not by default. Secure fields whose names identify them as API keys or secrets are omitted, along with interactive sign-in tokens. Each profile save starts with Save API keys & client secrets cleared, so the normal profile contains reusable settings without credentials.

If you explicitly select that option, credentials are added only after encryption with this app's Data Protection key. That file still contains sensitive material, can be restored only where the matching keys are available, and should not be shared.

Never. A session archive always excludes API keys, client secrets, and interactive authentication tokens. It can include non-secret settings, prompts, responses, metadata, and generated media so you can restore the work without packaging the credentials used to create it.

A profile is a compact JSON file for reusing playground settings. A session is a ZIP archive that combines a non-secret profile with the saved output and any supported media files. Use profiles to repeat a configuration and sessions to continue or review a complete run.

The default profile and every session omit credentials, which makes accidental key disclosure less likely. Still review files before sharing: endpoints, prompts, responses, metadata, uploaded content, and generated media may contain confidential or personal information. Never share a profile saved with encrypted credentials.

Choose Chat for conversations and reasoning, Image or Video for generation, Document AI for OCR and extraction, and Benchmark when you want to run the same task against two or three compatible models and compare or score the results.

Confirm that the endpoint belongs to the selected provider and deployment, the credential has access, and any required model or deployment name is correct. Also check regional availability, quota, content filters, and whether the endpoint expects a different authentication mode.

No. Estimates are a convenience based on the information available for the request and model. Your provider's metering, contract, region, and billing records remain the source of truth. The cost estimate shown in the playground excludes additional costs incurred through tool and MCP use.