To identify whether model input, model output, or expanded tool usage is driving the cost spike, you should focus on token usage.
Granular tracking of token usage breaks down the costs directly:
Input vs. Output Tokens: Tracing token usage isolates whether the user prompt (input) has grown or if the model is generating longer, more expensive responses (output).
Tool Calling: Expanded tool usage consumes extra tokens each time the agent schema processes an execution loop, which will clearly reflect as inflated token counts.
Reference:
https://www.mindstudio.ai/blog/subtraction-principle-agent-harness-optimization

Explanation:
Box 1: "{{$userId}}"
User Isolation (Scope)
Setting scope="{{$userId}}" instructs the Microsoft Foundry Agent Service to automatically map and partition memory to the logged-in individual using their Microsoft Entra user identity template.
The platform handles this dynamically from request headers, removing the need for manual client-side user ID management Box 2: [memory_tool] Tool List: The tools configuration parameter expects an iterable containing instantiated tool objects. Wrapping memory_tool inside a Python list ([memory_tool]) exposes the capability to the agent definition.
Reference:
https://learn.microsoft.com/en-us/azure/foundry/agents/how-to/memory-usage?pivots=python

The Azure AI Language service's extractive summarization feature is the appropriate choice to summarize video transcripts while preserving original sentence structure and order. Using the REST API or client library, you can set sentenceCount to exactly three to generate a concise summary from key phrases.
This approach avoids generating new, abstract text, fulfilling the constraint to keep original text, unlike abstractive summarization.
Key Implementation Details:
Service: Azure AI Language, specifically the Summarization feature.
Method: Extractive summarization selects the most relevant sentences directly from the source text, which is ideal for maintaining the original context, speaker attribution, and sentence structure.
Configuration: The API call allows setting the sentenceCount parameter to three, ensuring the output is limited to that exact number.
Workflow: This model works best on conversational transcripts processed via speech-to-text (e.g., Azure AI Speech), where speakers are identified, allowing the summarization tool to aggregate sentences while preserving the speaker-to-text relationship.
Reference:
https://learn.microsoft.com/en-us/azure/ai-services/language-service/summarization/how-to/text-summarization

The solution does not meet the goal. Protected material detection is intended to identify large language model output that matches known protected text or code, such as copyrighted text, selected web content, song lyrics, articles, recipes, or code. Microsoft describes protected material detection as a control for preventing AI- generated content from reproducing known protected material, not as a control for image safety or prompt injection.
The stated risk has two parts: users can upload unsafe images, and users can embed hidden instructions in images to manipulate the model. Unsafe image uploads require image moderation, because Azure AI Content Safety provides image APIs that detect harmful content across modalities and can support blocking decisions by harm category and severity. Hidden instructions extracted from images are indirect prompt injection or document attacks; Microsoft Prompt Shields are the capability designed to detect user prompt attacks and document attacks, including harmful instructions embedded in third-party content.
Therefore, protected material detection alone does not mitigate either primary risk. Reference topics: Azure AI Content Safety, image moderation, Prompt Shields, document attacks, indirect prompt injection, and protected material detection.
Topic 1, Case Study Contoso, Ltd
Overview
Contoso, Ltd is a multinational retail company that builds, deploys, and manages generative Al and agent- based solutions by using Microsoft Foundry.
Identity Environment:
Contoso uses Microsoft Entra ID for identity management, authentication, and authorization capabilities that enable agents to access organizational resources and services.
Contoso recently formed a new Al engineering team named Agent1Dev Team to optimize and maintain existing Al solutions.
The team collaborates with solution architects, DevOps engineers, and security engineers to design, implement, monitor, and secure Al applications.
Contoso also has a team named Agent1Test Team that is responsible for validating Al solutions before the solution deployments.
Generative Environment:
Contoso has a Microsoft Foundry deployment that contains two projects named Project1 and Project2.
Project1
Project1 contains a customer support agent named Agent1 that assists customers with product inquiries and troubleshooting requests.
Agent1 has the following configurations:
* Agent1 uses a base model deployment.
* A safety evaluation pipeline is NOT enabled.
* Tool invocation approval workflows are NOT enabled.
* Conversation memory constraints are NOT configured.
Agent1 interacts with customers by using digital support channels and answers general questions about Contoso products.
Project1 is deployed to an Azure region located in the European Union (EU).
Agent1Dev Team will use Project1 to optimize and maintain Agent1.
Project2
Project2 contains a deployed video generation model. The marketing department at Contoso has access to Project2 and plans to use the model to develop a video creation solution.
Development of the solution is incomplete.
Data Environment:
Contoso stores product-related information in Azure resources that support Al applications.
The Azure environment contains an Azure Blob Storage account named storage1 that stores product detail sheets for all the Contoso products.
The product sheets include specifications, feature descriptions, and product support information that Agent1 can use to answer customer questions. The product sheets are stored in the PDF format.
Problem Statement:
Contoso identifies the following issues:
* Agent1 has only general knowledge of the Contoso products.
* A recent chat interaction with Agent1 was analyzed for sentiment. The results of the analysis have NOT been processed yet.
* Agent1 does NOT use the detailed product information in the product sheets stored in storage1 when responding to customer questions.
* The finance department at Contoso reports that vendor invoices must be reviewed manually to ensure that the invoices match the terms defined in the vendor contracts. The invoices contain tables, logos, and varied layouts that make the documents difficult to process consistently.
Requirement:
Planned Changes:
Contoso plans to implement the following changes:
* Implement a solution for Project1 that analyzes the vendor invoices by evaluating both the visual layout and the textual content of the invoices, so that the invoice details can be verified against the vendor contract terms.
* Update the base model deployment used by Agent1 and standardize the model version to ensure continuity and consistent responses.
* Enable Agent1 to retrieve and use the detailed product information from the product sheets stored in storage1.
* Implement an indexing solution for the product sheets that Agent1 can use to answer customer questions.
* Complete the development of the video creation solution.
Technical Requirements:
Contoso identifies the following technical requirements:
* The model deployment used by Agent1 must support scalable, high-throughput generative Al workloads and dynamically scale to handle variable customer support traffic, without requiring reserved throughput capacity.
* The product sheets must be processed by using an indexing pipeline that enables semantic and vector search, so that Agent1 can retrieve the relevant product information.
* Responses generated by using the product sheet information must be relevant, complete, and accurate.
* Agent1 must be able to use the product sheets to answer natural language questions about product details.
* The model version used by Agent1 must remain consistent to ensure stable responses.
* The data processed by the model must remain within the EU.
Safety and Compliance Requirements:
Contoso identifies the following security and compliance requirements:
* API keys must NOT be used to access Foundry-deployed models.
* Access to the Azure resources must follow the principle of least privilege.
* The developers at Contoso must authenticate to Microsoft Foundry resources by using Microsoft Entra authentication.
* Access to Project1 must be assigned to the members of Agent1Dev Team by using a security group named SC_Agent1_Dev.
* Access to Project1 must be assigned to the members of Agent1Test Team by using a security group named SC_Agent1_Test.
* Agent1 must never reveal customer information, even if a document that contains customer data is added erroneously to the product sheet repository in storage1.
* The product sheets might contain images that include embedded text. Agent1 must be protected from malicious instructions potentially hidden within the images.
Business Information:
Contoso identifies the following business requirements:
* Users that interact with Agent1 must have a personalized experience in future interactions, including the ability for Agent1 to retain conversation context and recall relevant information from previous interactions.
* Agent1 must answer questions only about the products sold by Contoso.

Configure hybrid search , which executes full-text and vector queries within the same Azure AI Search request. The full-text component searches the product-name and product-code fields through the lexical index, providing the precision required for exact or near-exact identifiers. Microsoft specifically identifies product codes and other specialized terms as scenarios that frequently perform better with keyword search.
The vector component compares the embedding of the user's natural-language query with the embeddings stored for product descriptions. This retrieves semantically similar products even when the query and indexed description do not share the same literal words. Azure AI Search runs the full-text and vector searches in parallel and combines their result sets by using Reciprocal Rank Fusion, returning one unified ranking to the Foundry agent.
Keyword-only search would preserve exact matching but perform poorly for conceptual or paraphrased descriptions. Vector-only search supports semantic similarity but can miss precise product codes and rare identifiers. Semantic search alone reranks text-search results using language understanding; it does not replace the vector query required to use the existing embedding field.
Study Guide alignment: configure semantic, hybrid, and vector search for grounding, choose an appropriate retrieval method, and connect retrieval pipelines to agent tools .