Facing the error of capability host after using bicep code to create standard setup in AI foundry

Hi Balwant Singh

capabilityHost is the backing runtime that AI Foundry uses to wire up capabilities (model deployments, Content Understanding, evals, agents, etc.) to storage, Key Vault, logging, and registry.

• It is created at hub/project provisioning time only if you:
  1. Deploy the dependent resources with the hub (Storage, Key Vault, Log Analytics/App Insights, Container Registry), and
  2. Provide the required linkages (userOwnedStorage / data plane identities) at creation (not later), and
  3. Use a supported region for the specific capability set.

If any of those are missing or mis‑matched, the control plane returns:

1. Dependent resources were not provisioned with the hub

Your “standard setup” Bicep likely creates the hub (and maybe the project) but skips the “deps” module.

Fix

• Re-deploy the hub using the full quickstart pattern: Storage + Key Vault + Log Analytics/App Insights + Container Registry before or as part of the hub.
• Pass those resource IDs into the hub at creation.

2. BYO storage / default model linkages were omitted at creation

Several Foundry capabilities (e.g., Content Understanding GA) require resource-level defaults and/or user-owned storage at creation time. If you try to bind them after the hub exists, you won’t get a capabilityHost.

Fix

• Include the userOwnedStorage block (and other defaults) in the hub properties on initial deployment.
• If you’re using Content Understanding GA, ensure the model-defaults (e.g., gpt-4.1, gpt-4.1-mini, text-embedding-3-large) are set for the Foundry account before creating analyzers.

3. Region constraints

Even if eastus2 is a valid Foundry region, specific capabilities may not be fully enabled there. Creating the hub in one region and the deps in another can also block capabilityHost creation.

Fix

• Keep hub/project and all deps in the same supported region.
• If the capability you need is not enabled in eastus2, use a region where it is supported (e.g., eastus, westus3) or remove that capability until enabled.
• • capabilityHost is the backing runtime that AI Foundry uses to wire up capabilities (model deployments, Content Understanding, evals, agents, etc.) to storage, Key Vault, logging, and registry.
  • It is created at hub/project provisioning time only if you:
    1. Deploy the dependent resources with the hub (Storage, Key Vault, Log Analytics/App Insights, Container Registry), and
    2. Provide the required linkages (userOwnedStorage / data plane identities) at creation (not later), and
  • Use a supported region for the specific capability set.

    References:

    • Capability Hosts Overview
    • Troubleshooting Capability Hosts
    • Creating and Managing Hubs and Projects
    • Azure Deployment Operations