上下文基础

Context Grounding in Autopilot for Everyone allows you to search existing Context Grounding indexes to answer user queries.

For example, you want to query the HR documents of your organization, such as time and expense policies, or employee handbooks, using Autopilot. Using Context Grounding, an admin would create the index and enable it in Autopilot, which allows Autopilot to search the index to answer HR-related queries.

To use Context Grounding, make sure you meet the following criteria:

• 相关 PDF、CSV、JSON、DOCX、XLSX 和 TXT 文档上传到共享的 Orchestrator 存储桶中。
• 已在 Orchestrator 存储桶中创建索引。

Managing Context Grounding indexes

Index creation is done in Orchestrator, in the tenant-level Indexes page. For details, refer to Indexes.

Index enablement and management in Autopilot is done via the Context Grounding configuration section in Automation Cloud™ > Admin > AI Trust Layer > Autopilot for Everyone.

Once the index is created, enable it in Autopilot through the Context Grounding, panel.

Managing access for Context Grounding indexes

Access to Context Grounding indexes is managed at the folder level. Users can access indexes under the following conditions:

• 文件夹中存在索引
• 已将用户分配到索引所在的文件夹
• The index is enabled in Autopilot

Adding Context Grounding indexes

To add a Context Grounding index, follow the steps described in Creating indexes, in the Orchestrator user guide.

Enabling Context Grounding indexes

To enable an index for Autopilot for Everyone:

1. Navigate to the Automation Cloud > Admin > AI Trust Layer > Autopilot for Everyone tab, then choose the tenant where Autopilot for Everyone is installed.

2. Expand the Context grounding settings section.

3. Select the Enable Index option. The Enable Index In Autopilot panel opens.

4. From the Orchestrator folder dropdown, select the Orchestrator folder where your index resides.

5. From the Index name dropdown menu, select the index you want to add to Autopilot. The list of available indexes in the selected tenant is displayed.

6. In the Description for Autopilot* field, write a comprehensive description that should help Autopilot understand when to search for the selected index.

   例如，对于索引值“HR 文档”，其说明可以为“关于时间/费用、父子关系、PTO 的人力资源策略文档，和员工手册”。 用于回答有关公司人力资源政策的问题，为员工提供准确的最新信息。

7. For indexes with structured tabular data, select the Enable tabular query checkbox to allow retrieval of that information.

8. Select Enable to add the index. The panel closes and you are returned to the Context Grounding table.

重要提示：

系统会自动保存更改。

Editing Context Grounding indexes

To edit an index for Autopilot for Everyone:

1. Navigate to the Automation Cloud > Admin > AI Trust Layer > Autopilot for Everyone tab, then choose the tenant where Autopilot for Everyone is installed.
2. Expand the Context grounding settings section.
3. For the desired index, select Edit. The Updating index opens.
4. 根据需要修改所需的属性。
5. Select Save to update the index. The panel closes and you are returned to the Context Grounding table.

重要提示：

系统会自动保存更改。

Deleting Context Grounding indexes

To remove an index from Autopilot for Everyone:

1. Navigate to the Automation Cloud > Admin > AI Trust Layer > Autopilot for Everyone tab, then choose the tenant where Autopilot for Everyone is installed.
2. Expand the Context grounding settings section.
3. For the desired index, select Delete. A confirmation message is displayed.
4. Select Delete to delete the prompt, or Cancel to dismiss the action.

重要提示：

系统会自动保存更改。

将表格查询与上下文基础一起使用

本页说明如何为上下文基础索引启用功能，并分享准备数据和编写查询的最佳实践。

关于表格查询功能

The tabular query feature helps Autopilot for Everyone understand and answer natural language questions using structured data, such as CSV files. With it, you can query data, run comparisons, and get insights from your indexed tables directly in chat.

启用表格查询

To enable the feature, check the Enable tabular query box when setting up a Context Grounding index in Autopilot for Everyone.

启用后，Autopilot 可以使用索引文件中的结构化数据回答自然语言问题。

准备数据最佳实践

您的结果取决于源数据的整洁程度和格式良好程度。使用以下提示准备和优化数据，以便更准确地查询：

确保数据类型正确无误

Numerical operations, such as comparisons or calculations, need correct data types to work. Autopilot for Everyone needs to recognize numbers as numeric types (Real or Integer), not as text.

要检查 Autopilot 已分配给列的数据类型，请使用以下提示：

• “列出 <index_name> 中的 5 行数据，并列出所有字段及其数据类型。”

确保数字列和百分比列保持一致

在创建索引之前，从数据列中删除特殊字符，例如百分号 %。Autopilot 将20%等值视为文本，这会阻止数值查询正常工作。

Make sure date columns use a consistent format, preferably MM/DD/YYYY. Use the data type check query to confirm that date columns are parsed as Date Time rather than plain text.

表格查询最佳实践

清晰的提示会带来更好的结果。查询表格数据时，请记住以下提示：

避免复杂的架构

要小心包含许多列的文件。数据架构越复杂，自然语言查询被曲解的可能性就越大。

注意 null 值

计算会忽略查询列中带有 null 或空值的行。例如，当您计算四列的总和时，系统会排除其中一列具有空值的任何行。为避免这种情况，请规范化您的数据并将 Null 替换为适当的值。

指定字符串匹配类型

默认情况下，查询引擎使用精确匹配来比较字符串。要查找部分匹配项，您必须将其显式包含在查询中：

• For an exact match: "BCA Claims and Consulting Ltd"
• For a partial match: "BCA Claims and Consulting Ltd. or any of its subsidiaries". This query returns any result that contains the substring "BCA Claims and Consulting Ltd."

使用调试视图

对于高级故障排除，请使用Ctrl+Alt+Shift+D打开调试视图。该面板显示查询引擎如何解释您的自然语言提示，这可以帮助您诊断意外结果。