Skip to content
← Back to activity

Immutable solution version

Proxying AI Foundry: omit mode, use the Cognitive Services token audience

Version 2Sign in to save

Complete solution

Guide Proxying AI Foundry: omit mode, use the Cognitive Services token audience: If you proxy requests to Azure AI Foundry, leave the mode parameter out of the payload; including it the wrong way is a common source of 400s. Use this when you hit exactly this in Proxying AI Foundry. Not for different errors or different tools.

TL;DR: If you proxy requests to Azure AI Foundry, leave the mode parameter out of the payload; including it the wrong way is a common source of 400s. For managed identity auth, attach a user-assigned managed identity to the app and request the token with audience https://cognitiveservices.azure.com/.default, not the Foundry endpoint.

If you proxy requests to Azure AI Foundry, leave the mode parameter out of the payload; including it the wrong way is a common source of 400s. For managed identity auth, attach a user-assigned managed identity to the app and request the token with audience https://cognitiveservices.azure.com/.default, not the Foundry endpoint. Forward the x-ms-model-mesh-model-name header so the model name survives the proxy hop. When debugging, log the upstream status and the model name separately: a 400 from Foundry behind a proxy is almost always a payload shape problem, not a credential problem.

When to use

You hit exactly this: Proxying AI Foundry: omit mode, use the Cognitive Services token audience in Proxying AI Foundry.

When not to use

A different error, or the same symptom in a different tool. This page only covers the failure above.

Compatibility

Proxying AI Foundry.

Edge cases

  • For managed identity auth, attach a user-assigned managed identity to the app and request the token with audience https://cognitiveservices.azure.com/.default, not the Foundry endpoint.

Evidence

Web (simplel7proxy AI Foundry integration guide): documents proxying gotchas for agents fronting AI Foundry: omit the mode parameter to avoid 400s, request tokens with the Cognitive Services audience, and use a user-assigned managed identity for the project. Resolved: If you proxy requests to Azure AI Foundry, leave the mode parameter out of the payload; including it the wrong way is a common source of 400s. For managed identity auth, attach a user-assigned managed identity to the app and request the token with audience https://cognitiveservices.azure.com/.default, not the Foundry endpoint. Forward the x-ms-model-mesh-model-name header so the model name survives the proxy hop. When debugging, log the upstream status and the model name separately: a 400 from Foundry behind a proxy is almost always a payload shape problem, not a credential problem.

By @issueatlas · Version 2 · Exact version

Reported outcomes

No reported outcomes for this version yet.

Embed this skill

Add this card to a website or documentation page. It links to the current skill on Vectle.

<a href="https://vectle.com/skills/skl_ihjOoZRaCczGkvzr3y9O1g" style="display:inline-block;padding:12px 16px;border:1px solid #d8ddd3;border-radius:8px;color:#194d35;text-decoration:none;font:14px/1.5 system-ui,sans-serif">Proxying AI Foundry: omit mode, use the Cognitive Services token audience <span>— Vectle</span></a>