feat: integrate Agent SDK into kernel, migrate to GlobalStandard model types, remove Greeting function, and update Orchestrator to handle greetings #613

Vamshi-Microsoft · 2025-07-18T11:23:17Z

Purpose

This pull request introduces multiple changes across documentation, infrastructure, scripts, and backend code to enhance functionality, improve clarity, and update configurations. Key updates include adjustments to AI model capacities, documentation improvements for reusing existing resources, and the addition of a SQL query generator agent. Below is a categorized summary of the most important changes:

AI Model Capacity Updates:

Reduced GPT_MIN_CAPACITY from 250 to 200 and increased TEXT_EMBEDDING_MIN_CAPACITY from 40 to 80 in .github/workflows/CAdeploy.yml.
Updated default capacities for gpt-4o-mini to 200 and text-embedding-ada-002 to 80 in docs/QuotaCheck.md and infra/scripts/quota_check_params.sh. [1] [2] [3]

Documentation Enhancements:

Added a detailed guide for reusing an existing Log Analytics Workspace in docs/re-use-log-analytics.md and linked it in other documentation files. [1] [2] [3]

Infrastructure Updates:

Changed the SKU for the embedding model from Standard to GlobalStandard in infra/deploy_ai_foundry.bicep and corresponding JSON files. [1] [2] [3]
Updated retry mechanism for file uploads in infra/scripts/copy_kb_files.sh to increase retries from 3 to 5 and added exponential backoff with detailed logging.

Backend Enhancements:

Added a new SQL query generator agent in src/App/backend/agents/agent_factory.py to generate T-SQL queries based on user input and predefined rules.
Removed the greeting function from src/App/backend/plugins/chat_with_data_plugin.py as conversational responses are now handled by updated agent instructions.

These changes collectively improve deployment configurations, enhance user guidance, optimize infrastructure, and expand backend capabilities.

Does this introduce a breaking change?

Yes
No

Golden Path Validation

I have tested the primary workflows (the "golden path") to ensure they function correctly without errors.

Deployment Validation

I have validated the deployment process successfully and all services are running as expected with this change.

What to Check

Verify that the following are valid

...

Other Information

…nt to handle Greetings

…nt to handle greetings

…inputs

* updated model capacity minimume to 200 (#602) * docs: Added file and images (#577) * Create exp.md * Add files via upload * Delete docs/images/re_use_log/exp.md * Add files via upload * Update CustomizingAzdParameters.md * Update DeploymentGuide.md add the section Reusing an Existing Log Analytics Workspace * Update re-use-log-analytics.md update the back link url and remove extra space --------- Co-authored-by: Roopan-Microsoft <[email protected]> Co-authored-by: Thanusree-Microsoft <[email protected]> * docs: Add and update links for CustomizingAzdParameters.md and re-use-log-analytics.md (#606) * Update CustomizingAzdParameters.md Added link * Update re-use-log-analytics.md Updated link * fix: Increase retry attempts and improve error messaging for Azure Blob Storage uploads (#607) --------- Co-authored-by: Priyanka-Microsoft <[email protected]> Co-authored-by: Atulku-Microsoft <[email protected]> Co-authored-by: Roopan-Microsoft <[email protected]> Co-authored-by: Thanusree-Microsoft <[email protected]>

fix: Update agent instructions to restrict web search conversational inputs

feat: migrate model types to GlobalStandard

Copilot

Pull Request Overview

This pull request introduces a major architectural refactoring that integrates the Agent SDK into the kernel, migrates to GlobalStandard model types for Azure OpenAI deployments, removes the greeting function from the ChatWithDataPlugin, and updates the orchestrator to handle conversational interactions through enhanced agent instructions.

Key changes include:

Migration from direct OpenAI calls to Agent SDK for SQL query generation
Infrastructure updates to use GlobalStandard SKU for text-embedding models
Removal of dedicated greeting functionality in favor of enhanced agent instructions

Reviewed Changes

Copilot reviewed 15 out of 18 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/App/backend/plugins/chat_with_data_plugin.py	Removed greeting function and migrated SQL query generation to use Agent SDK
src/App/backend/agents/agent_factory.py	Added new SQL agent factory method with enhanced instructions for conversational handling
src/App/tests/backend/plugins/test_chat_with_data_plugin.py	Updated tests to reflect Agent SDK migration and removed greeting tests
infra/deploy_ai_foundry.bicep	Changed embedding model SKU from Standard to GlobalStandard
docs/re-use-log-analytics.md	Added new documentation for reusing existing Log Analytics workspaces

Comments suppressed due to low confidence (1)

src/App/tests/backend/plugins/test_chat_with_data_plugin.py:153

The test should verify that mock_get_connection was called to ensure the error handling path is properly tested when the agent succeeds but database connection fails.

        assert "Error retrieving SQL data" in result

Copilot · 2025-07-18T11:24:15Z

src/App/backend/agents/agent_factory.py

+                agent_instructions = config.SQL_SYSTEM_PROMPT or """
+    You are an expert assistant in generating T-SQL queries based on user questions.
+    Always use the following schema:
+    1. Table: Clients (ClientId, Client, Email, Occupation, MaritalStatus, Dependents)
+    2. Table: InvestmentGoals (ClientId, InvestmentGoal)
+    3. Table: Assets (ClientId, AssetDate, Investment, ROI, Revenue, AssetType)
+    4. Table: ClientSummaries (ClientId, ClientSummary)
+    5. Table: InvestmentGoalsDetails (ClientId, InvestmentGoal, TargetAmount, Contribution)
+    6. Table: Retirement (ClientId, StatusDate, RetirementGoalProgress, EducationGoalProgress)
+    7. Table: ClientMeetings (ClientId, ConversationId, Title, StartTime, EndTime, Advisor, ClientEmail)
+
+    Rules:
+    - Always filter by ClientId = <provided>
+    - Do not use client name for filtering
+    - Assets table contains snapshots by date; do not sum values across dates
+    - Use StartTime for time-based filtering (meetings)
+    - Only return the raw T-SQL query. No explanations or comments.
+    """
+


[nitpick] The fallback instructions should be stored as a constant rather than embedded as a multi-line string. This improves maintainability and makes the instructions easier to update.

Suggested change

agent_instructions = config.SQL_SYSTEM_PROMPT or """

You are an expert assistant in generating T-SQL queries based on user questions.

Always use the following schema:

1. Table: Clients (ClientId, Client, Email, Occupation, MaritalStatus, Dependents)

2. Table: InvestmentGoals (ClientId, InvestmentGoal)

3. Table: Assets (ClientId, AssetDate, Investment, ROI, Revenue, AssetType)

4. Table: ClientSummaries (ClientId, ClientSummary)

5. Table: InvestmentGoalsDetails (ClientId, InvestmentGoal, TargetAmount, Contribution)

6. Table: Retirement (ClientId, StatusDate, RetirementGoalProgress, EducationGoalProgress)

7. Table: ClientMeetings (ClientId, ConversationId, Title, StartTime, EndTime, Advisor, ClientEmail)

Rules:

- Always filter by ClientId = <provided>

- Do not use client name for filtering

- Assets table contains snapshots by date; do not sum values across dates

- Use StartTime for time-based filtering (meetings)

- Only return the raw T-SQL query. No explanations or comments.

"""

agent_instructions = config.SQL_SYSTEM_PROMPT or SQL_AGENT_FALLBACK_PROMPT

Copilot · 2025-07-18T11:24:16Z

src/App/backend/agents/agent_factory.py

+                    name="SQLQueryGeneratorAgent",
+                )
+
+                cls._sql_agent = {"agent": agent, "client": project_client}


The singleton pattern implementation lacks proper cleanup. Consider adding a cleanup method similar to delete_all_agent_instance for the SQL agent to prevent resource leaks.

Copilot · 2025-07-18T11:24:16Z

infra/scripts/copy_kb_files.sh

 az storage blob upload-batch --account-name "$storageAccount" --destination data/"$extractedFolder1" --source $extractionPath1 --auth-mode login --pattern '*' --overwrite --output none
 if [ $? -ne 0 ]; then
-    retries=3
+    maxRetries=5
+    retries=$maxRetries
    sleepTime=10
-    echo "Error: Failed to upload files to Azure Blob Storage. Retrying upload...($((4 - retries)) of 3)"
+    attempt=1
    while [ $retries -gt 0 ]; do
+        echo "Error: Failed to upload files to Azure Blob Storage. Retrying upload...$attempt of $maxRetries in $sleepTime seconds"
        sleep $sleepTime
        az storage blob upload-batch --account-name "$storageAccount" --destination data/"$extractedFolder1" --source $extractionPath1 --auth-mode login --pattern '*' --overwrite --output none
        if [ $? -eq 0 ]; then
            echo "Files uploaded successfully to Azure Blob Storage."
            break
        else
            ((retries--))
-            echo "Retrying upload... ($((4 - retries)) of 3)"
+            ((attempt++))
            sleepTime=$((sleepTime * 2))
-            sleep $sleepTime
        fi
    done
-    exit 1
+    if [ $retries -eq 0 ]; then
+        echo "Error: Failed to upload files after all retry attempts."
+        exit 1
+    fi
 else
    echo "Files uploaded successfully to Azure Blob Storage."
 fi

 az storage blob upload-batch --account-name "$storageAccount" --destination data/"$extractedFolder2" --source $extractionPath2 --auth-mode login --pattern '*' --overwrite --output none
 if [ $? -ne 0 ]; then
-    retries=3
+    maxRetries=5
+    retries=$maxRetries
+    attempt=1
    sleepTime=10
-    echo "Error: Failed to upload files to Azure Blob Storage. Retrying upload...($((4 - retries)) of 3)"
    while [ $retries -gt 0 ]; do
+        echo "Error: Failed to upload files to Azure Blob Storage. Retrying upload...$attempt of $maxRetries in $sleepTime seconds"
        sleep $sleepTime
        az storage blob upload-batch --account-name "$storageAccount" --destination data/"$extractedFolder2" --source $extractionPath2 --auth-mode login --pattern '*' --overwrite --output none
        if [ $? -eq 0 ]; then
            echo "Files uploaded successfully to Azure Blob Storage."
            break
        else
            ((retries--))
-            echo "Retrying upload... ($((4 - retries)) of 3)"
+            ((attempt++))
            sleepTime=$((sleepTime * 2))
-            sleep $sleepTime
        fi
    done
-    exit 1
+    if [ $retries -eq 0 ]; then
+        echo "Error: Failed to upload files after all retry attempts."
+        exit 1
+    fi
 else
    echo "Files uploaded successfully to Azure Blob Storage."
 fi


The retry logic duplicates the same code block twice with only minor variations. Consider extracting this into a reusable function to reduce code duplication.

github-actions · 2025-07-22T09:12:56Z

🎉 This PR is included in version 1.7.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀

Vamshi-Microsoft and others added 19 commits July 15, 2025 10:24

refactor: Remove greeting kernal function and Update orchestrator age…

37fe076

…nt to handle Greetings

Remove greeting test

7e0dc8b

Removing unnecessary decorators

17719fe

Resolved PyLint issue

587d902

Resolve PyLint issue

1ac5db0

refactor: Remove greeting kernel function and Update orchestrator age…

50dd6f3

…nt to handle greetings

integrated Agent SDK into the SQL Kernel function

f8f77c5

resolved pylint

e695659

updated test file

496dc3e

updated test file

15ce8aa

updated test file

f3a14eb

feat: integrate agent sdk into the kernel function

71c4df6

fix: Update agent instructions to restrict web search conversational …

161bfb4

…inputs

Merge branch 'dev' into psl-AgentInstruction

f670185

Updated test file

a582bfe

Merge pull request #610 from microsoft/psl-AgentInstruction

5893ea4

fix: Update agent instructions to restrict web search conversational inputs

updated standard to globalstandard

f0d0cda

Merge pull request #611 from microsoft/update-standard-to-globalstandard

9b5931e

feat: migrate model types to GlobalStandard

Copilot AI review requested due to automatic review settings July 18, 2025 11:23

Vamshi-Microsoft requested review from Avijit-Microsoft, Prajwal-Microsoft, Roopan-Microsoft, Vinay-Microsoft and aniaroramsft as code owners July 18, 2025 11:23

Copilot AI reviewed Jul 18, 2025

View reviewed changes

Avijit-Microsoft approved these changes Jul 22, 2025

View reviewed changes

Avijit-Microsoft merged commit b684d25 into main Jul 22, 2025
15 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: integrate Agent SDK into kernel, migrate to GlobalStandard model types, remove Greeting function, and update Orchestrator to handle greetings #613

feat: integrate Agent SDK into kernel, migrate to GlobalStandard model types, remove Greeting function, and update Orchestrator to handle greetings #613

Uh oh!

Vamshi-Microsoft commented Jul 18, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jul 18, 2025

Uh oh!

Copilot AI Jul 18, 2025

Uh oh!

Copilot AI Jul 18, 2025

Uh oh!

Uh oh!

github-actions bot commented Jul 22, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

feat: integrate Agent SDK into kernel, migrate to GlobalStandard model types, remove Greeting function, and update Orchestrator to handle greetings #613

feat: integrate Agent SDK into kernel, migrate to GlobalStandard model types, remove Greeting function, and update Orchestrator to handle greetings #613

Uh oh!

Conversation

Vamshi-Microsoft commented Jul 18, 2025

Purpose

AI Model Capacity Updates:

Documentation Enhancements:

Infrastructure Updates:

Backend Enhancements:

Does this introduce a breaking change?

Golden Path Validation

Deployment Validation

What to Check

Other Information

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI Jul 18, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jul 18, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jul 18, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Jul 22, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants