Create and manage agents

Create an agent

Create an agent from the Agent admin page in the Snowsight UI to answer questions and provide insights by following the steps in Configure and interact with Agents.

The agent uses a semantic view, semantic model, a Cortex Search service, or a combination of these to provide answers. Although you can create a basic agent that doesn’t use any of these tools, that basic agent exclusively uses the base model to provide answers. As a result, the agent lacks access to data within your Snowflake account and has limited context for answers.

The semantic view or model is used for tables and structured data in Snowflake. This data can be in any table that is queryable by Snowflake. Cortex Search is useful for indexing and searching pieces of data with a lot of text, such as documents, conversations, and transcripts.

Modify an existing agent

For instructions on modifying the configuration for an existing agent, including adding tools and updating other details, see Add tools.

Use the agent

After you’ve created an agent, you can ask it questions to get insights from your data. The agent can answer questions such as the following:

  • What is the average sales amount for the last quarter?

  • What product sold the most units last month?

  • Can you show me the sales trend for the last year?

It can also provide the following visualizations:

  • Bar chart

  • Line chart

  • Pie chart

  • Scatter plot

Note

Snowflake recommends that you use separate browser tabs for Snowsight and Snowflake Intelligence so that you can switch between them.

To use the agent, follow these steps:

  1. From the left-hand navigation, select AI & ML or navigate to the Snowflake Intelligence landing page.

  2. Select Snowflake Intelligence. The navigation changes to focus on Snowflake Intelligence. We recommend opening a new browser tab specifically for Snowsight while using the Snowflake Intelligence application. By using distinct browser tabs, you can switch between the Snowsight interface and the Snowflake Intelligence application.

  3. In the chat window, select the agent from the dropdown.

  4. Select the data source for the agent to use.

  5. Ask the agent a question or ask it to provide you with a visualization of your data.

Use research mode

By default, Snowflake Intelligence agents balance speed and quality when answering questions.

If you have a complex question or want the agent to explore all possible options, in the chat window, select the Enable Research mode. The agent will be more thorough, but the process might take more time. This setting remains selected for the duration of the current session.

Zero-setup file upload

Snowflake Intelligence supports uploading files directly in the chat interface to provide additional context for the agent. The agent can use the content of the files to answer questions and provide insights.

When you upload a file, it is automatically saved in your user stage. The file is only accessible to you and the agent during your session. The file does not persist across sessions. If you want to use the file in future sessions, you must move it to a permanent location, such as a table or stage.

Snowflake Intelligence supports the following file types for zero-setup file upload:

  • CSV

  • XLSX

  • PDF

  • PPTX

Each file must be less than 50 MB, but you can upload up to 5 files.

If the document requires complex processing, the user’s default warehouse can be used to run Snowpark code for the agent to better analyze and process the data.

Observability

You can view logs for an agent to see details about the interactions that users have had with the agent. The logs include information such as the prompts that users have sent to the agent, the responses that the agent has provided, and any errors that have occurred. For more information about viewing logs for agents, see Monitor Cortex Agent requests.

When users in your organization interact with agents, they can provide feedback about the responses that the agents provide. This feedback gives high-level insights about the satisfaction of users. To view user feedback for your agents,see Monitor Cortex Agent requests.

Language: English