Snowflake Intelligence 概述¶

使用 Snowflake Intelligence 获取见解并根据组织中的数据采取行动。借助 Snowflake Intelligence，您可以执行以下操作：

使用自然语言创建图表并即时获得答案。您无需技术专业知识或等待自定义仪表板即可发现趋势和分析数据。
访问和分析数千个数据源，包括结构化数据和非结构化数据。您可以同时连接来自电子表格、文档、图像和数据库的见解。

Snowflake Intelligence uses agents, which are AI models that are connected to one or more semantic views, semantic models, Cortex search services, and tools. Agents can answer questions, provide insights, and show visualizations. Snowflake Intelligence is powered by Cortex AI Functions, Cortex Analyst, and Cortex Search.

根据以下章节设置 Snowflake Intelligence，并开始使用它来从数据中获取价值。有关 Snowflake Intelligence 的快速入门指南，请参阅 Snowflake Intelligence 入门 (https://quickstarts.snowflake.com/guide/getting-started-with-snowflake-intelligence/index.html)。

支持的模型和区域¶

Snowflake Intelligence 支持以下模型。只要账户有权访问这些模型，您就可以使用它们。有关更多信息，请参阅控制模型访问。

Claude 4.5
Claude 4.0
Claude 3.7
Claude 3.5
GPT 5
GPT 4.1

虽然列出的模型可能并非在所有区域都可用，但您可以借助 Cortex 跨区域推理在任何云或区域中使用 Snowflake Intelligence。这包括模型不可用的云和区域。有关配置 Cortex 跨区域推理的更多信息，请参阅跨区域推理。

When creating an agent, we recommend selecting Auto for the model. This lets Snowflake Intelligence automatically select the highest quality model for your account and automatically improves as new models become available.

AWS US - In AWS, Claude 4+ offers the highest quality and best speed performance. We recommend that you set up Cortex Cross-region inference for aws_us to use Claude 4 and get the best performance. Without Cortex Cross-region inference, you are restricted to using Claude 3.5 in aws_us.
Azure US - If you are using Snowflake Intelligence in East US, you can use GPT 4.1+ without Cortex Cross-region inference. Other region and model combinations require Cortex Cross-region inference setup for azure_us.
AWS EU - You can use Claude 4+ in this region as long as you configure Cortex Cross-region inference for aws_eu.
AWS APJ - You can use Claude 4+ in this region as long as you configure Cortex Cross-region inference for aws_apj.

设置 Snowflake Intelligence¶

To set up Snowflake Intelligence for your users, do the following:

备注

The ACCOUNTADMIN role is the only role that has the CREATE SNOWFLAKE INTELLIGENCE ON ACCOUNT privilege required to create a Snowflake Intelligence object.

Create a Snowflake Intelligence object. The Snowflake Intelligence object is a single object meant to manage all agents used with Snowflake Intelligence in your account. You can only have one Snowflake Intelligence object in your account.
Add agents to the Snowflake Intelligence object.
GRANT USAGE privileges to the Snowflake Intelligence object.

Create a Snowflake Intelligence object¶

You can either use the Snowflake Intelligence UI or SQL to create a Snowflake Intelligence object.

Snowflake automatically creates the Snowflake Intelligence object when you modify the Snowflake Intelligence settings for the first time. When created using the UI, the Snowflake Intelligence object is named SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT. You can't specify a different name.

Sign in to Snowsight.

In the navigation menu, select AI & ML » Agents.

Select the Snowflake Intelligence tab.

Select Open settings. The Snowflake Intelligence object is created automatically if it doesn't already exist. You can then add agents to the object.
To create a Snowflake Intelligence object, you can use the following command:
CREATE SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT;
Copy

Add agents¶

The Snowflake Intelligence object is an account-level object that contains a list of agents. You can add or remove agents from this object to create a curated list of agents for your users. For more information about adding or removing agents, see Configure the visibility of agents in Snowflake Intelligence.

Grant Snowflake Intelligence privileges¶

The following privileges control access to Snowflake Intelligence objects:

CREATE SNOWFLAKE INTELLIGENCE ON ACCOUNT: Account-level privilege that allows creating a Snowflake Intelligence object. This privilege is granted to ACCOUNTADMIN by default.

To grant this privilege to another role, run the following command:
```
GRANT CREATE SNOWFLAKE INTELLIGENCE ON ACCOUNT TO ROLE <role_name>;
```
Copy
USAGE: Object-level privilege that allows users to view the list of agents added to the Snowflake Intelligence object and see configuration values.

To grant this privilege, run the following command:
```
GRANT USAGE ON SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT TO ROLE <role_name>;
```
Copy
ALTER: Object-level privilege that allows users to add or remove agents from the Snowflake Intelligence object and change configuration values. Account administrators have this privilege by default.

To grant this privilege, run the following command:
```
GRANT MODIFY ON SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT TO ROLE <role_name>;
```
Copy

To make the Snowflake Intelligence object visible to all of your users, grant USAGE privileges on the object to the PUBLIC role:

GRANT USAGE ON SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT TO ROLE PUBLIC;

Copy

As an ADMIN, you also have ALTER privileges on the Snowflake Intelligence object. This allows you to add or remove agents from the object to create a curated list of agents for your users.

To set up Snowflake Intelligence for your users, you must configure agent privileges. For information about the privileges required for agents, see 访问控制要求.

重要

默认情况下，Snowflake Intelligence 使用用户的默认角色和默认仓库。当您邀请其他人使用 Snowflake Intelligence 时，请确保他们已设置了默认角色和仓库。

备注

来自 Snowflake Intelligence 的所有查询都使用用户的凭据。与用户关联的所有基于角色的访问控制和数据掩码策略都会自动应用于与代理的所有互动和对话。

创建代理¶

To get started, create an agent that users can interact with in Snowflake Intelligence. For information about creating an agent, see 创建代理. For best practices when creating an agent, see Best Practices to Building Cortex Agents.

Configure the visibility of agents in Snowflake Intelligence¶

If you haven't created a Snowflake Intelligence object and added agents to it, users see all agents they have access to in your account.

For the optimal experience, create a curated list of agents by adding them to the Snowflake Intelligence object. This allows you to control which agents appear in the Snowflake Intelligence interface for all users.

To see whether the Snowflake Intelligence object has been created in your account, use the following command:

SHOW SNOWFLAKE INTELLIGENCES;

Copy

备注

Only one Snowflake Intelligence object can exist in an account.

Managing agents with the Snowflake Intelligence object¶

To add agents to the Snowflake Intelligence object, use the following command:

ALTER SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT ADD AGENT <db.schema.agent_name>;

Copy

To remove agents from the Snowflake Intelligence object, use the following command:

ALTER SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT DROP AGENT <db.schema.agent_name>;

Copy

备注

Any user or admin with the right database and schema privileges can create agents. However, agents are not automatically added to the Snowflake Intelligence object. Users must have the ALTER privilege on the Snowflake Intelligence object and USAGE privileges on the agent to add an agent to the Snowflake Intelligence object.

Administrators must have the USAGE privilege on the agent to add it to the Snowflake Intelligence object.

Agent visibility logic¶

Snowflake Intelligence uses the following logic to determine which agents are visible to users:

If the Snowflake Intelligence object exists and contains one or more agents, users see the individual agents from this curated list that they have access.
If the Snowflake Intelligence object exists but contains zero agents AND the SNOWFLAKE_INTELLIGENCE.AGENTS schema exists with one or more agents, users see agents from the SNOWFLAKE_INTELLIGENCE.AGENTS schema. This ensures that Snowflake Intelligence doesn't break if you haven't migrated your agents yet.
If the Snowflake Intelligence object doesn't exist or contains zero agents AND the SNOWFLAKE_INTELLIGENCE.AGENTS schema doesn't exist or contains zero agents, users see all agents that they have access to in the account.

Migrating from managing agent visibility with the SNOWFLAKE_INTELLIGENCE.AGENTS schema¶

重要

The SNOWFLAKE_INTELLIGENCE.AGENTS schema is deprecated as a mechanism for managing agent visibility. If you're currently using this schema, we recommend migrating to the Snowflake Intelligence object.

If you're using the SNOWFLAKE_INTELLIGENCE.AGENTS schema, your agents will continue to work, as detailed in Agent visibility logic. However, migrating to the Snowflake Intelligence object provides the following benefits:

Flexibility: Create and manage agents anywhere in your account without needing to centralize them in a single schema.

Improved permission management: Separate the ability to create agents from the ability to publish them in Snowflake Intelligence.

Avoid naming conflicts: Eliminate potential conflicts with the SNOWFLAKE_INTELLIGENCE.AGENTS schema name.

Easier agent visibility management: Use a single object to control which agents appear to all users.

Make sure you've created a Snowflake Intelligence object before you migrate your agents. For information about creating a Snowflake Intelligence object, see 设置 Snowflake Intelligence.

After you've created an object, use the following code to add an agent to the Snowflake Intelligence object:

ALTER SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT ADD AGENT SNOWFLAKE_INTELLIGENCE.AGENTS.<agent_name>;

Copy

You can rename or drop the SNOWFLAKE_INTELLIGENCE database or AGENTS schema after the migration. For more information, see ALTER SCHEMA.

Configure Snowflake Intelligence with private connectivity¶

Snowflake Intelligence supports integration with AWS Privatelink and Azure Private Link to establish a private connection between your Virtual Private Cloud (VPC) or virtual network (VNet) and Snowflake Intelligence. Configuring private connectivity requires setting up the correct DNS resolution to direct traffic to the Snowflake Intelligence service through this private connection.

Note that AWS PrivateLink and Azure Private Link are not services provided by Snowflake. They are an AWS service and Microsoft service, respectively, that Snowflake supports to use with your Snowflake account.

先决条件¶

Complete the following prerequisites before connecting to Snowflake Intelligence with private connectivity.

Set up AWS PrivateLink by following the instructions in AWS PrivateLink 和 Snowflake or Azure Private Link by following the instructions in Azure 专用链接和 Snowflake.
Ensure that a regionless-snowsight-privatelink-url is available by calling the SYSTEM$GET_PRIVATELINK_CONFIG function using the ACCOUNTADMIN system role.

重要

Snowflake Intelligence exclusively uses the regionless URL format for private connectivity access. Unlike other private connectivity URLs used for Snowflake, you should not include a region identifier, such as us-west-2, in the hostname. Any attempts to connect using a region-specific URL will result in a failure.

Connect to Snowflake Intelligence¶

Connect to Snowflake Intelligence by configuring the DNS for Snowflake Intelligence to use the subdomain.

Create a CNAME record in your private DNS zone, privatelink.snowflakecomputing.cn, that maps the following URL to the DNS name of your VPC or VNET endpoint.
```
si-<org-acct>.privatelink.snowflakecomputing.cn
```
Copy

After the configuration is complete, users within your network can access Snowflake Intelligence by navigating to the following URL:

https://si-<org-acct>.privatelink.snowflakecomputing.cn

Copy

The connection is routed securely over the private connection.

User authentication with private connectivity¶

Users accessing Snowflake Intelligence with private connectivity use the standard Snowflake authentication process, which requires them to provide their account identifier, username, and password on the login page.

Use the Snowflake-managed MCP server to connect to your agents¶

Any agent that you create in Snowflake, or the tools that the agent is connected to, can have a managed endpoint for other systems to connect with MCP. This provides a seamless integration layer for tools like Claude Desktop, Langgraph, and other tools that integrate with MCP.

The MCP server provides a standards-based interface that allows AI agents to discover and invoke tools, such as Cortex Analyst and Cortex Search, and retrieve the data they need. For more information, see Snowflake 管理的 MCP 服务器.

Update Snowflake Intelligence settings¶

You can modify settings for the Snowflake Intelligence interface that users interact with Cortex Agents through.

Sign in to Snowsight.
In the navigation menu, select AI & ML » Agents.
Select the Snowflake Intelligence tab.
Select Open settings.
From the General settings section, you can modify the following settings:
- Brand display name: The name of the Snowflake Intelligence interface that is displayed to users.
- Welcome message: The message that is displayed when users first open the Snowflake Intelligence interface.
- Primary domain: The domain that users use to access Snowflake Intelligence.
选择 Save。
From the Custom appearance section, you can modify the following settings:
- Color theme: The color theme of the Snowflake Intelligence interface. You can provide a custom primary color in hexadecimal format.
- Full-length logo and Compact logo: The logos that are displayed when the navigation pane is expanded or collapsed, respectively.
- Favicon: The icon that is displayed in the browser tab.
选择 Save。

故障排除¶

如果遇到 table / search service / stage does not exist 错误，则可能表明存在权限问题。验证以下权限设置是否正确：

对于每个语义模型：
- 用户的默认角色会获授语义模型暂存区或视图以及表的数据库和架构的 USAGE。
- If using the older semantic model, the user's default role is granted READ on the stage that stores the semantic model file.
- If using a semantic view, the user's default role is granted REFERENCES on the semantic view.
- 用户的默认角色会获授在语义模型或视图中定义的每个表的 SELECT。
对于每项 Cortex Search Service：
- 用户的默认角色会获授 Cortex Search Service 的数据库和架构的 USAGE。
- 用户会获授 Cortex Search Service 的 USAGE。

Legal notices¶

Where your configuration of Snowflake Intelligence uses a model provided on the Model and Service Flow-down Terms, your use of that model is further subject to the terms for that model on that page.

The data classification of inputs and outputs are as set forth in the following table.

Input data classification	Output data classification	Designation
Usage Data	Customer Data	Covered AI Features [1]

For additional information, refer to Snowflake AI 和 ML.