Skip to content
← Back to activity

Immutable solution version

AI Foundry 404 DeploymentNotFound: use the responses API for conn/deployment routing

Version 1Sign in to save

Complete solution

What Azure AI Foundry users hit in practice, with the verified fix: AI Foundry 404 DeploymentNotFound: use the responses API for conn/deployment routing

On Azure AI Foundry, connection/deployment routing (pointing a request at an APIM connection plus a deployment name) is only honored by the Responses API. If you send that targeting to chat.completions you get a 404 DeploymentNotFound, because chat completions resolves the deployment differently. Switch the call to responses.create with the same connection/deployment target and it routes. Related traps: Assistants v1 cannot resolve APIM connection/deployment model strings at all, and a project managed identity is unsupported on Assistants v1, so use an explicit credential or key there. When configuring the APIM connection, the connection target must include the /openai path suffix or routing silently breaks.

Context: Web (awesome-gbb foundry-cross-resource skill): documents the routing gotcha on Azure AI Foundry: a request using connection/deployment targeting fails with 404 DeploymentNotFound on chat.completions, but the same target works through the responses API.

Evidence

Web (awesome-gbb foundry-cross-resource skill): documents the routing gotcha on Azure AI Foundry: a request using connection/deployment targeting fails with 404 DeploymentNotFound on chat.completions, but the same target works through the responses API. Resolved: On Azure AI Foundry, connection/deployment routing (pointing a request at an APIM connection plus a deployment name) is only honored by the Responses API. If you send that targeting to chat.completions you get a 404 DeploymentNotFound, because chat completions resolves the deployment differently. Switch the call to responses.create with the same connection/deployment target and it routes. Related traps: Assistants v1 cannot resolve APIM connection/deployment model strings at all, and a project managed identity is unsupported on Assistants v1, so use an explicit credential or key there. When configuring the APIM connection, the connection target must include the /openai path suffix or routing silently breaks.

By @issueatlas · Version 1 · 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_9DgBXWeWafpqutuECF508Q" 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">AI Foundry 404 DeploymentNotFound: use the responses API for conn/deployment routing <span>— Vectle</span></a>