This provider targets the new Azure AI Foundry Account — an Microsoft.CognitiveServices/accounts resource of kind AIServices that hosts one or more Foundry Projects containing agents, evaluations, and model deployments.
Which Foundry is this? Polysync ships two distinct Foundry providers because Microsoft ships two SKUs that share the "AI Foundry" name. Pick the one that matches your Azure resource:
Azure AI Foundry Project (this provider) Azure AI Foundry (legacy) Azure resource provider Microsoft.CognitiveServices/accounts(kindAIServices)Microsoft.MachineLearningServices/workspacesConcept Foundry Account + Project AML Hub / Workspace Use this for Agents, evaluations, model deployment inference on the current-generation Foundry portal Pipeline / Command / Sweep jobs on AML Hub workspaces Job types Foundry Agent Run, Foundry Evaluation Run, Foundry Model Deployment Inference AI Foundry Pipeline Job, Command Job, Sweep Job Required role Foundry User (formerly Azure AI User) on the account, plus Reader for control-plane discovery Contributor on the workspace Control-plane API 2024-10-01(CognitiveServices)2024-04-01(MachineLearningServices)Projects data plane https://{account}.services.ai.azure.com/api/projects/{project}(api-versionv1)n/a API Key auth supported Yes — inference only No Icon Blue hexagonal flask Purple brain/circuit tile If you see
AuthorizationFailedonMicrosoft.MachineLearningServices/workspaces/{name}while using a CognitiveServices Foundry resource, you have selected the wrong provider — switch to this one.
polysync-foundry-dev).api-key header). Execution-only and limited to Model Deployment Inference. Agent and Evaluation jobs require an AAD identity. Because key-based platforms cannot list deployments (no control plane), Get Jobs is empty — create the inference job manually: Job Type Foundry Model Deployment Inference, External Id and Deployment Name set to the deployment, Inference Mode set to chat or embedding, and any instruction in the job's System Prompt setting. (Adding a user_prompt task parameter to a manually created job is currently affected by a save defect.)Polysync Service Principal stores no credentials — it authenticates as the multi-tenant Polysync enterprise application. The Setup editor shows that application's (client) ID beside the auth-method hint: grant that application the role below on the resource.
az provider register --namespace <provider-namespace> --wait.Networking. Polysync calls the resource's own endpoint, so allow-list Polysync's outbound address (shown in the Setup editor) on any firewall. A resource restricted to a private endpoint cannot be reached — see Networking and firewalls on the Platform page.
Foundry User (formerly Azure AI User — same role id, renamed by Microsoft) on the Foundry account — the agents / projects data plane. Assign at the project scope to limit the identity to one project.
Reader on the account — control-plane discovery (Microsoft.CognitiveServices/accounts/deployments/read); already covered by Contributor/Owner. API Key auth never uses the control plane.
Cognitive Services OpenAI User on the account — model deployment inference (chat/completions, embeddings) for AAD methods. API Key auth presents the account key instead.
Validated live 2026-09-30 with an identity holding exactly those three roles: connectivity, control-plane discovery, agent discovery/run, and chat + embedding inference all passed.
| Plane | Base URL | Token scope |
|---|---|---|
| Control plane | https://management.azure.com/.../Microsoft.CognitiveServices/accounts/{account} |
https://management.azure.com/.default |
| Projects data plane (agents, evaluations) | https://{account}.services.ai.azure.com/api/projects/{project} |
https://ai.azure.com/.default |
| Model deployments data plane | https://{account}.services.ai.azure.com/openai/deployments/{deployment} |
https://cognitiveservices.azure.com/.default |
Azure AI Foundry Project exposes three Polysync job types. See each dedicated page for parameter handling, output binding, execution flow, monitor URL, and troubleshooting:
https://ai.azure.com/.default; model deployments use https://cognitiveservices.azure.com/.default.AuthorizationFailed for the MachineLearningServices resource provider — you are using the wrong provider. This provider targets Microsoft.CognitiveServices/accounts. For AML Hub workspaces use Azure AI Foundry instead.