mirror of
https://github.com/microsoft/semantic-kernel.git
synced 2026-10-12 02:19:06 +00:00
Evan Mattson
fb0e418a7c
### Motivation and Context
The AzureAIAgent streaming function call invocation is broken when
handling more than one required function call from the service. There is
nested logic to handle passing the tool calls back to the service
because it requires one to pass in a handler.
In this PR, we're now controlling everything from the main loop and are
passing back the handler as the event stream for further processing.
This does require some logic to determine if we have an initial async
context manager (based on the first call to create the run stream) or if
we're now using the handler from submitting tool calls. We can now see
that when we have multiple functions involved, we are able to get the
results as we expect:
```
Sample Output:
# User: 'What is the price of the special drink and then special food item added together?'
Function Call:> MenuPlugin-get_specials with arguments: {}
Function Result:>
Special Soup: Clam Chowder
Special Salad: Cobb Salad
Special Drink: Chai Tea
for function: MenuPlugin-get_specials
Function Call:> MenuPlugin-get_item_price with arguments: {"menu_item": "Chai Tea"}
Function Call:> MenuPlugin-get_item_price with arguments: {"menu_item": "Clam Chowder"}
Function Result:> $9.99 for function: MenuPlugin-get_item_price
Function Result:> $9.99 for function: MenuPlugin-get_item_price
Function Call:> MathPlugin-Add with arguments: {"input":9.99,"amount":9.99}
Function Result:> 19.98 for function: MathPlugin-Add
# AuthorRole.ASSISTANT: The price of the special drink, Chai Tea, is $9.99 and the price of the special food
item, Clam Chowder, is $9.99. Added together, the total price is $19.98.
```
<!-- Thank you for your contribution to the semantic-kernel repo!
Please help reviewers and future users, providing the following
information:
1. Why is this change required?
2. What problem does it solve?
3. What scenario does it contribute to?
4. If it fixes an open issue, please link to the issue here.
-->
### Description
Fix handling multiple function calls for the streaming AzureAIAgent.
- Fixes a missing `quote` from the url citation for streaming annotation
content
- Shows the correct way to retrieve Bing Grounding tool calls in the two
concept samples.
- Adds handling for tool content content for OpenAPI plugins.
- Adds logging for streaming agent invocation tool calls.
- Relaxes the handling of the `FunctionResultContent` id attribute to be
`str | None` similar to that of `FunctionCallContent`. The type hints
for `id` have always been `str | None`, but the actual type hint on the
attribute has been `str`. We don't always get a `FunctionResultContent`
id from the Azure Agent Service during streaming invocations - message
addition via the overide `__add__` method still work properly.
- Move the AzureAIAgent pkg dependencies to be installed by default,
instead of requiring the `azure` extra.
- Closes #12312
- Closes #12328
- Closes #12331
- Closes #12324
<!-- Describe your changes, the overall approach, the underlying design.
These notes will help understanding how your code works. Thanks! -->
### Contribution Checklist
<!-- Before submitting this PR, please make sure: -->
- [X] The code builds clean without any errors or warnings
- [X] The PR follows the [SK Contribution
Guidelines](https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
and the [pre-submission formatting
script](https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md#development-scripts)
raises no violations
- [X] All unit tests pass, and I have added new tests where possible
- [X] I didn't break anyone 😄
Semantic Kernel Concepts by Feature
Table of Contents
Agents - Creating and using agents in Semantic Kernel
Azure AI Agent
- Azure AI Agent as Kernel Function
- Azure AI Agent with Auto Function Invocation Filter Streaming
- Azure AI Agent with Auto Function Invocation Filter
- Azure AI Agent with Azure AI Search
- Azure AI Agent with Bing Grounding Streaming with Message Callback
- Azure AI Agent with Bing Grounding
- Azure AI Agent with Code Interpreter Streaming with Message Callback
- Azure AI Agent Declarative with Azure AI Search
- Azure AI Agent Declarative with Bing Grounding
- Azure AI Agent Declarative with Code Interpreter
- Azure AI Agent Declarative with File Search
- Azure AI Agent Declarative with Function Calling From File
- Azure AI Agent Declarative with OpenAPI Interpreter
- Azure AI Agent Declarative with Existing Agent ID
- Azure AI Agent File Manipulation
- Azure AI Agent Prompt Templating
- Azure AI Agent Message Callback Streaming
- Azure AI Agent Message Callback
- Azure AI Agent Streaming
- Azure AI Agent Structured Outputs
- Azure AI Agent Truncation Strategy
Bedrock Agent
- Bedrock Agent Simple Chat Streaming
- Bedrock Agent Simple Chat
- Bedrock Agent With Code Interpreter Streaming
- Bedrock Agent With Code Interpreter
- Bedrock Agent With Kernel Function Simple
- Bedrock Agent With Kernel Function Streaming
- Bedrock Agent With Kernel Function
- Bedrock Agent Mixed Chat Agents Streaming
- Bedrock Agent Mixed Chat Agents
Chat Completion Agent
- Chat Completion Agent as Kernel Function
- Chat Completion Agent Function Termination
- Chat Completion Agent Templating
- Chat Completion Agent Message Callback Streaming
- Chat Completion Agent Message Callback
- Chat Completion Agent Summary History Reducer Agent Chat
- Chat Completion Agent Summary History Reducer Single Agent
- Chat Completion Agent Truncate History Reducer Agent Chat
- Chat Completion Agent Truncate History Reducer Single Agent
Mixed Agent Group Chat
- Mixed Chat Agents Plugins
- Mixed Chat Agents
- Mixed Chat Files
- Mixed Chat Images
- Mixed Chat Reset
- Mixed Chat Streaming
OpenAI Assistant Agent
- Azure OpenAI Assistant Declarative Code Interpreter
- Azure OpenAI Assistant Declarative File Search
- Azure OpenAI Assistant Declarative Function Calling From File
- Azure OpenAI Assistant Declarative Templating
- Azure OpenAI Assistant Declarative With Existing Agent ID
- OpenAI Assistant Auto Function Invocation Filter Streaming
- OpenAI Assistant Auto Function Invocation Filter
- OpenAI Assistant Chart Maker Streaming
- OpenAI Assistant Chart Maker
- OpenAI Assistant Declarative Code Interpreter
- OpenAI Assistant Declarative File Search
- OpenAI Assistant Declarative Function Calling From File
- OpenAI Assistant Declarative Templating
- OpenAI Assistant Declarative With Existing Agent ID
- OpenAI Assistant File Manipulation Streaming
- OpenAI Assistant File Manipulation
- OpenAI Assistant Retrieval
- OpenAI Assistant Message Callback Streaming
- OpenAI Assistant Message Callback
- OpenAI Assistant Streaming
- OpenAI Assistant Structured Outputs
- OpenAI Assistant Templating Streaming
- OpenAI Assistant Vision Streaming
OpenAI Responses Agent
- Azure OpenAI Responses Agent Declarative File Search
- Azure OpenAI Responses Agent Declarative Function Calling From File
- Azure OpenAI Responses Agent Declarative Templating
- OpenAI Responses Agent Declarative File Search
- OpenAI Responses Agent Declarative Function Calling From File
- OpenAI Responses Agent Declarative Web Search
- OpenAI Responses Message Callback Streaming
- OpenAI Responses Message Callback
- OpenAI Responses File Search Streaming
- OpenAI Responses Plugins Streaming
- OpenAI Responses Reuse Existing Thread ID
- OpenAI Responses Web Search Streaming
Audio - Using services that support audio-to-text and text-to-audio conversion
- Chat with Audio Input
- Chat with Audio Output
- Chat with Audio Input and Output
- Audio Player
- Audio Recorder
AutoFunctionCalling - Using Auto Function Calling to allow function call capable models to invoke Kernel Functions automatically
- Azure Python Code Interpreter Function Calling
- Function Calling with Required Type
- Parallel Function Calling
- Chat Completion with Auto Function Calling Streaming
- Functions Defined in JSON Prompt
- Chat Completion with Manual Function Calling Streaming
- Functions Defined in YAML Prompt
- Chat Completion with Auto Function Calling
- Chat Completion with Manual Function Calling
- Nexus Raven
ChatCompletion - Using ChatCompletion messaging capable service with models
- Simple Chatbot
- Simple Chatbot Kernel Function
- Simple Chatbot Logit Bias
- Simple Chatbot Store Metadata
- Simple Chatbot Streaming
- Simple Chatbot with Image
- Simple Chatbot with Summary History Reducer Keeping Function Content
- Simple Chatbot with Summary History Reducer
- Simple Chatbot with Truncation History Reducer
- Simple Chatbot with Summary History Reducer using Auto Reduce
- Simple Chatbot with Truncation History Reducer using Auto Reduce
ChatHistory - Using and serializing the ChatHistory
Filtering - Creating and using Filters
- Auto Function Invoke Filters
- Function Invocation Filters
- Function Invocation Filters Stream
- Prompt Filters
- Retry with Filters
Functions - Invoking Method or Prompt functions with Kernel
Grounding - An example of how to perform LLM grounding
Local Models - Using the OpenAI connector and OnnxGenAI connector to talk to models hosted locally in Ollama, OnnxGenAI, and LM Studio
- ONNX Chat Completion
- LM Studio Text Embedding
- LM Studio Chat Completion
- ONNX Phi3 Vision Completion
- Ollama Chat Completion
- ONNX Text Completion
Logging - Showing how to set up logging
Memory - Using Memory AI concepts
- Simple Memory
- Memory Data Models
- Memory with Pandas Dataframes
- Complex memory
- Full sample with Azure AI Search including function calling
Model-as-a-Service - Using models deployed as serverless APIs on Azure AI Studio to benchmark model performance against open-source datasets
On Your Data - Examples of using AzureOpenAI On Your Data
- Azure Chat GPT with Data API
- Azure Chat GPT with Data API Function Calling
- Azure Chat GPT with Data API Vector Search
Plugins - Different ways of creating and using Plugins
- Azure Key Vault Settings
- Azure Python Code Interpreter
- OpenAI Function Calling with Custom Plugin
- Plugins from Directory
Processes - Examples of using the Process Framework
PromptTemplates - Using Templates with parametrization for Prompt rendering
- Template Language
- Azure Chat GPT API Jinja2
- Load YAML Prompt
- Azure Chat GPT API Handlebars
- Configuring Prompts
RAG - Different ways of RAG (Retrieval-Augmented Generation)
Reasoning - Using ChatCompletion to reason with OpenAI Reasoning
Search - Using Search services information
Service Selector - Shows how to create and use a custom service selector class
Setup - How to set up environment variables for Semantic Kernel
Structured Outputs - How to leverage OpenAI's json_schema Structured Outputs functionality
TextGeneration - Using TextGeneration capable service with models
Configuring the Kernel
In Semantic Kernel for Python, we leverage Pydantic Settings to manage configurations for AI and Memory Connectors, among other components. Here’s a clear guide on how to configure your settings effectively:
Steps for Configuration
-
Reading Environment Variables:
- Primary Source: Pydantic first attempts to read the required settings from environment variables.
-
Using a .env File:
- Fallback Source: If the required environment variables are not set, Pydantic will look for a
.envfile in the current working directory. - Custom Path (Optional): You can specify an alternative path for the
.envfile viaenv_file_path. This can be either a relative or an absolute path.
- Fallback Source: If the required environment variables are not set, Pydantic will look for a
-
Direct Constructor Input:
- As an alternative to environment variables and
.envfiles, you can pass the required settings directly through the constructor of the AI Connector or Memory Connector.
- As an alternative to environment variables and
Microsoft Entra Token Authentication
To authenticate to your Azure resources using a Microsoft Entra Authentication Token, the AzureChatCompletion AI Service connector now supports this as a built-in feature. If you do not provide an API key -- either through an environment variable, a .env file, or the constructor -- and you also do not provide a custom AsyncAzureOpenAI client, an ad_token, or an ad_token_provider, the AzureChatCompletion connector will attempt to retrieve a token using the DefaultAzureCredential.
To successfully retrieve and use the Entra Auth Token, you need the Cognitive Services OpenAI Contributor role assigned to your Azure OpenAI resource. By default, the https://cognitiveservices.azure.com token endpoint is used. You can override this endpoint by setting an environment variable .env variable as AZURE_OPENAI_TOKEN_ENDPOINT or by passing a new value to the AzureChatCompletion constructor as part of the AzureOpenAISettings.
Best Practices
- .env File Placement: We highly recommend placing the
.envfile in thesemantic-kernel/pythonroot directory. This is a common practice when developing in the Semantic Kernel repository.
By following these guidelines, you can ensure that your settings for various components are configured correctly, enabling seamless functionality and integration of Semantic Kernel in your Python projects.