Domain 2 of 3 · Chapter 14 of 15

Configure custom prompts to use the Foundry model catalog

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing Designing and Building Integrated AI Agent Solutions in Copilot Studio premium course — no separate purchase.

14-day money-back guarantee — no questions asked.

Included in this chapter:

  • Two catalogs behind one Model field
  • What the Foundry side must produce first
  • Connecting the deployment, field by field
  • The endpoint suffix is where this fails
  • What the feature supports, and what it refuses
  • Selecting the model, and what that selection binds
  • Image input narrows the Model dropdown
  • Where safety and governance live now
  • How this shows up in questions

The two catalogs behind a prompt's Model field

PropertyManaged model (built-in catalog)Connected Microsoft Foundry model
How it reaches the Model fieldAlready listed; you select itModel field, plus sign, Connect a model from Azure AI Foundry, then Connect
What you supplyNothing beyond the selectionModel deployment name and Base model name exactly as in Foundry, plus the Azure model endpoint URL
Who deployed the modelMicrosoft, as part of the productYour organization, in a Foundry resource, before the prompt can use it
Published discriminatorsCategory (Mini, General, Deep), Licensing (Basic, Standard, Premium rate), release-stage tag, regionWhatever the deployment is: base model, version, capacity type, content filtering configuration, rate limits
Supported model typesThe generative models the catalog lists for promptsChat completion type only; the endpoint URL must end in /chat/completions
Content moderation level sliderAvailable; Low to High, default ModerateUnavailable; the setting is available only for managed models
Where safety is then configuredThe prompt's own Content moderation level settingThe deployment's content filtering configuration and Responsible AI policies managed in Foundry
Platform governance handleThe prompt and agent settings in Copilot StudioThe Azure AI Foundry connector and the connection it creates, governed by data policy

Decision tree

Does the thing you want to attachhave its own instructions and tools?YesConnected Foundry agentAgents page, Agent Idnot the Model fieldNoIs the deployment achat completion model?NoThis route is closedembeddings, image generationREST API tool or managed modelYesIs it in the GPT-5 familyor a later generation?YesNot supported herebring your own model excludes ituse the managed catalogNoDoes the endpoint URL endin /chat/completions?NoFix the suffix firsta Responses API URL fails withResource not foundYesConnect itModel deployment name andBase model name, exactly as in FoundryThen select it on the prompteach run of that prompt uses the selected modelthe moderation slider is gone: filter in Foundry
Cheat sheet

Cheat sheet

  • Connect catalog models through a Prompt tool
  • Enter exact deployment and base-model names
  • Use a chat-completions endpoint
  • Select the connected model on the prompt
  • Treat primary and feature-specific models as separate selections
  • Filter models by input capability
  • Govern Foundry model connections with data policy
  • Use supported chat-completion models
  • Apply model-appropriate safety governance

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. Change the model version and settings - Microsoft Copilot Studio
  2. Bring your own model for your prompts - Microsoft Copilot Studio
  3. Endpoints for Microsoft Foundry Models
  4. Deploy Microsoft Foundry Models in the Foundry portal
  5. Create a prompt - Microsoft Copilot Studio
  6. Select a primary AI model for your agent - Microsoft Copilot Studio
  7. Add inputs to your prompt - Microsoft Copilot Studio
  8. Responsible AI for Microsoft Foundry
  9. Data policies - Power Platform