Create declarative agents with Microsoft 365 Copilot Tuning models (preview)

Microsoft 365 Copilot Tuning allows organizations to fine-tune large language models (LLMs) by using their own tenant data. These fine-tuned models power agents that can perform domain-specific tasks based on the organization's unique knowledge. This article explains how to create declarative agents in Copilot Studio agent builder that use those fine-tuned models.

Prerequisites

To create a declarative agent based on a fine-tuned model, you need:

• A Microsoft 365 production environment with a Microsoft 365 Copilot license.
• Access to a fine-tuned model. The model maker assigns security groups to the model when they create it.

Create the agent

To create the agent:

1. In Microsoft 365 Copilot Chat, select Create agent in the left-hand navigation to open Copilot Studio agent builder.

2. In the What is the purpose of your agent? dialog box, select Task-specific, and then choose Continue.

   Note

   If you don't see the dialog box asking the agent's purpose, Copilot Tuning might not be enabled in your organization, or you might not have access to any fine-tuned models. Check with your organization's administrators.

3. In the next dialog box, select a fine-tuned model from the list of models, then select Done.

4. Finish customizing your agent by filling in the remaining fields. For more information, see Build agents with Copilot Studio agent builder.

Add knowledge sources

Agents that use a model tuned for expert question & answer (Q&A) support adding SharePoint or OneDrive knowledge sources. Other knowledge sources aren't supported.

Agents that use a model tuned for document generation or document summarization don't support adding knowledge sources.

Related content

• Microsoft 365 Copilot Tuning overview
• Build agents with Copilot Studio agent builder