Copilot Studio

A Microsoft Copilot Studio agent’s default orchestration model is Microsoft’s and cannot be swapped for an arbitrary endpoint. There are three ways to bring governance in, from most-supported to workaround.

See Connect your harness for your base URL, key, and model IDs.

Copilot Studio supports a custom model deployed in Azure AI Foundry, both inside a Prompt tool and as the agent’s Primary response model (Settings → Generative AI → Primary response model → Connect a model).

1. In Azure AI Foundry, deploy a model from the catalog as a Chat Completion deployment. Note its Deployment name, Base model name, Endpoint URI, and API key.
2. In Copilot Studio, open the model connection dialog and click Connect a model from Azure AI Foundry.
3. Enter the Deployment name and Base model name exactly as in Foundry, the Endpoint URI (must end in /chat/completions), and the API key.
4. Validate the connection, then Connect.

This governs through Foundry, not TapPass. Use it when your model lives in Foundry.

The Foundry connection expects an OpenAI-compatible endpoint ending in /chat/completions, which TapPass also serves. Mechanically you may be able to paste https://app.tappass.ai/v1/chat/completions and a TapPass key instead of a real Foundry endpoint.

Copilot Studio supports MCP servers and connectors. Registering TapPass’s governed tools governs what the agent does (tool calls, data access), which is often the better fit, since the action side carries the most risk. The model can stay Microsoft’s while every action runs through TapPass policy and audit.

For full model control, build a custom engine agent in Copilot Studio and point its orchestrator at TapPass.

For model-plane connections, send one message and confirm a governed event in the TapPass console. For tool-plane connections, trigger a tool call and confirm the tool event is governed.