Create a new agent

Agents powered by the GitHub Copilot harness in Microsoft Copilot Studio use instructions written in natural language, knowledge, tools, and other components to reason over context and take action. In this article, you manually create an agent, save it, and optionally update its primary language, solution, and schema name.

You can also create agents and workflows by using the agentic builder (preview), which suggests and builds solutions by using a maker's guidance. Learn more in the quickstart for the agentic builder.

1. Sign in to Copilot Studio.

2. On the Home page, select the Agent tile.

   Alternatively, select Agents in the side pane to go to the Agents page, and then select New Agent.

   Your agent opens in the agent designer with the Build tab active and the name field in focus.

3. Enter a name for your agent in the name field.

4. Optionally, change the primary language, solution, and schema name under Agent Details in Settings before you save your agent.

   Note

   The last agent name you enter before the first save determines the schema name for your agent. Then, this unique system name used to reference the agent becomes read-only.

5. In the Instructions section, describe how the agent should behave, including its role, tone, and any rules or guidelines it must follow. Learn more in Configure the agent's details and instructions.

6. Select the Save icon.

   If you skipped entering a name for your agent, a message appears prompting you for the name:

   1. Enter the name you want.
   2. Select Save.

7. After saving, you can continue to build your agent:

   • Add knowledge sources. Knowledge is the data your agent can access when answering user questions. Learn more about how to add knowledge to an agent.
   • Add tools. Tools are external capabilities your agent can invoke during a conversation. Learn more about how to add tools to an agent.
   • Add skills. Skills are reusable capabilities that extend what your agent can do. Learn more about how to add skills to an agent.
   • Add connected agents. Connected agents are agents your primary agent can invoke during a conversation. Learn more about how to add connected agents to an agent.
   • Configure the model for your agent.
   • Turn on and configure memory for your agent.

8. After building, you can also test your agent and share it with others. Learn more in Preview and test an agent, Evaluate an agent, and Share your agent with others.

9. Once your agent is ready, you can publish it to make it available for use and monitor its performance. Learn more in Publish and distribute and Monitor your agent.

Next steps

• Configure the agent's details and instructions