Revising content

Author: HeidiSteen

articles/search/media/agentic-retrieval/agentic-retrieval-architecture.png

@@ -16,9 +16,9 @@ ms.date: 05/08/2025
 
 [!INCLUDE [Feature preview](./includes/previews/preview-generic.md)]
 
-In Azure AI Search, *agentic retrieval* is a new parallel query processing architecture that uses conversational language models to generate multiple subqueries for a single retrieval request, incorporating conversation history and semantic ranking to produce high-quality grounding data for custom chat and generative AI solutions.
+In Azure AI Search, *agentic retrieval* is a new parallel query processing architecture that uses conversational language models to generate multiple subqueries for a single retrieval request, incorporating conversation history and semantic ranking to produce high-quality grounding data for custom chat and generative AI solutions that include agents.
 
-Programmatically, agentic retrieval is supported through a new Agents object in the 2025-05-01-preview data plane REST API and in Azure SDK prerelease packages that provide the feature. An agent's retrieval response is designed for downstream consumption by other agents and chat apps based on generative AI.
+Programmatically, agentic retrieval is supported through a new Knowledge Agents object (also known as a search agent) in the 2025-05-01-preview data plane REST API and in Azure SDK prerelease packages that provide the feature. An agent's retrieval response is designed for downstream consumption by other agents and chat apps based on generative AI.
 
 ## Why use agentic retrieval
 
@@ -47,15 +47,15 @@ Agentic retrieval invokes the entire query processing pipeline multiple times fo
 
 ## Agentic retrieval architecture
 
-Agentic retrieval is designed for a conversational search experience that includes an LLM. An important part of agentic retrieval is that an entire chat conversation can be included as inputs in subsequent queries, providing context and nuance for more relevant responses.
+Agentic retrieval is designed for a conversational search experience that includes an LLM. An important part of agentic retrieval is how the LLM breaks down an initial query into subqueries, which are more effective at locating the best matches in your index.
 
 :::image type="content" source="media/agentic-retrieval/agentic-retrieval-architecture.png" alt-text="Diagram of agentic retrieval workflow using an example query." lightbox="media/agentic-retrieval/agentic-retrieval-architecture.png" :::
 
 Agentic retrieval has these components:
 
 | Component | Resource | Usage |
 |-----------|----------|-------|
-| LLM (gpt-4o and gpt-4.1 series) | Azure OpenAI | Formulates subqueries for the query plan. You can use these models for other downstream operations. Specifically, you can send the unified response string to one of these models and ask it ground its answer on the string. |
+| LLM (gpt-4o and gpt-4.1 series) | Azure OpenAI | An LLM has two functions. First, it formulates subqueries for the query plan and sends it back to the search agent. Second, after the query executes, the LLM receives grounding data from the query response and uses it for answer formulation. |
 | Search index | Azure AI Search | Contains plain text and vector content, a semantic configuration, and other elements as needed. |
 | Search agent | Azure AI Search | Connects to your LLM, providing parameters and inputs to build a query plan. |
 | Retrieval engine | Azure AI Search | Executes on the LLM-generated query plan and other parameters, returning a rich response that includes content and query plan metadata. Queries are keyword, vector, and hybrid. Results are merged and ranked. |
@@ -139,7 +139,7 @@ Putting it all together, you'd pay about $3.30 for semantic ranking in Azure AI
 
 ## How to get started
 
-You must use the REST APIs or a prerelease Azure SDK page that provides the functionality. At this time, there's no Azure portal or Azure AI Foundry portal support.
+You must use the preview REST APIs or a prerelease Azure SDK package that provides the functionality. At this time, there's no Azure portal or Azure AI Foundry portal support.
 
 Choose any of these options for your next step.
 

articles/search/search-agentic-retrieval-concept.md

@@ -42,7 +42,7 @@ After the wizard completes, you have the following components:
 |-----------|----------|------------|
 | Search index | Azure AI Search | Contains indexed content from a supported Logic Apps connector. The index schema is a default index created by the wizard. You can add extra elements, such as scoring profile or semantic configuration, but you can't change existing fields. You view, manage, and access the search index on Azure AI Search. |
 | Logic app resource and workflow | Azure Logic Apps | You can view the running workflow, or you can open the designer in Azure Logic Apps to edit the workflow, as you regularly do if you'd started from Azure Logic Apps instead. You can edit and extend the workflow, but exercise caution so as to not break the indexing pipeline. |
-| Logic app templates | Azure Logic Apps | Up to two templates created per workflow: one for on-demand indexing, and a second template for scheduled indexing. You can modify the indexing schedule. |
+| Logic app templates | Azure Logic Apps | Up to two templates created per workflow: one for on-demand indexing, and a second template for scheduled indexing. You can modify the indexing schedule in the **Index multiple documents** step of the workflow. |
 
 ## Prerequisites
 
@@ -96,7 +96,7 @@ The following connectors are helpful for indexing unstructured data, as a comple
 
 Currently, the public preview has these limitations:
 
-+ Search index is generated using a fixed schema (document ID, content, and vectorized content), with text extraction only. You can [modify the index](#modify-existing-objects) as long as the update doesn't affect existing fields.
++ The search index is generated using a fixed schema (document ID, content, and vectorized content), with text extraction only. You can [modify the index](#modify-existing-objects) as long as the update doesn't affect existing fields.
 + Vectorization supports text embedding only.
 + Deletion detection isn't supported. You must manually [delete orphaned documents](search-howto-reindex.md#delete-orphan-documents) from the index.
 + Duplicate documents in the search index are a known issue in this preview. Consider deleting objects and starting over if this becomes an issue.
@@ -113,8 +113,7 @@ Follow these steps to create a Logic Apps workflow for indexing content in Azure
 
 1. In **Connect to your data**, provide a name prefix used for the search index and workflow. Having a common name helps you manage them together.
 
-<!-- Open Issue: how to specify frequency -->
-1. Specify the indexing frequency. If you choose on a schedule, a template that includes scheduling option is created in Logic Apps.
+1. Specify the indexing frequency. If you choose on a schedule, a template that includes a scheduling option is used to create the workflow. You can modify the indexing schedule in the **Index multiple documents** step of the workflow after it's created.
 
 1. Select an authentication type where the logic app workflow connects to the search engine and starts the indexing process. The workflow can connect using  [Azure AI Search API keys](search-security-api-keys.md) or the wizard can create a role assignment that grants permissions to the Logic Apps system-assigned managed identity, assuming one exists.
 
@@ -142,9 +141,10 @@ You can make the following modifications to a search index without breaking inde
 
 You can make the following updates to a workflow without breaking indexing:
 
-+ Modify templates that control indexing frequency.
++ Modify **List files in folder** to change the number of documents sent to indexing.
 + Modify **Chunk Text** to vary token inputs. The recommended token size is 512 tokens for most scenarios.
 + Modify **Chunk Text** to add a page overlap length.
++ Modify **Index multiple documents** step to control indexing frequency if you chose scheduled indexing in the wizard.
 
 In logic apps designer, review the workflow and each step in the indexing pipeline. The workflow specifies document extraction, default document chunking ([Text Split skill](cognitive-search-skill-textsplit.md)), embedding ([Azure OpenAI embedding skill](cognitive-search-skill-azure-openai-embedding.md)), output field mappings, and finally indexing.