Skip to content

README.md

Latest commit

 

History

History
101 lines (73 loc) · 4.5 KB

README.md

File metadata and controls

101 lines (73 loc) · 4.5 KB

Sample: AI

A sample demonstrating various AI capabilities in the Python Teams SDK.

Prerequisites

  • Python 3.12 or later
  • UV package manager
  • An Microsoft 365 development account. If you don't have one, you can get one for free by signing up for the Microsoft 365 Developer Program.

Setup

  1. Install dependencies:
uv sync
  1. Set up your .env file with your Azure OpenAI API key (or OpenAI API Key):
AZURE_OPENAI_API_KEY=<your_azure_openai_api_key>
AZURE_OPENAI_ENDPOINT=<your_azure_openai_endpoint>
AZURE_OPENAI_MODEL=<your_azure_openai_model_deployment_name>
AZURE_OPENAI_API_VERSION=<your_azure_openai_api_version>

# Alternatively, set the OpenAI API key:
OPENAI_API_KEY=<sk-your_openai_api_key>

Run

# Activate virtual environment
source .venv/bin/activate  # On macOS/Linux
# .venv\Scripts\Activate   # On Windows

# Run the AI test
python examples/ai-test/src/main.py

Usage

From Teams, DevTools, or your test client, use any of the following commands to trigger specific scenarios:

Scenario Usage Description
Simple LLM check hi Basic ChatPrompt with pirate personality
Function calling pokemon <pokemon_name> Single function calling - searches Pokemon via PokeAPI
Multi-Function calling weather Multiple function calling - gets location then weather
Streaming stream <your query> Shows streaming responses in verbose language
Citations citations Demonstrates citation functionality with position markers
Model switching model completions / model responses Switch between Chat Completions and Responses API models
Plugin stats plugin Shows AI plugin function call statistics
Memory management memory clear Clears conversation memory
Feedback collection feedback demo Demonstrates message feedback with like/dislike buttons
Feedback statistics feedback stats <message_id> Shows feedback summary for a specific message
Stateful interactions <any other message> Shows persistent conversation memory across interactions

Features Demonstrated

Core AI Functionality

  • ChatPrompt - Basic LLM interaction patterns with optional persistent memory
  • String instructions - Simple instruction passing (vs SystemMessage objects)
  • Model switching - Runtime switching between different AI models

Function Calling

  • Single functions - Pokemon search with real API integration
  • Multiple functions - Location detection followed by weather lookup
  • Error handling - Proper exception handling for API failures

Advanced Features

  • Streaming responses - Real-time response streaming with group/1:1 handling
  • Memory management - Per-conversation memory with manual clearing
  • Custom plugins - AI plugin system with lifecycle hooks
  • Citations - Position-based citations with proper formatting
  • Feedback collection - Message feedback with like/dislike reactions and text feedback

Best Practices

  • AI-generated indicators - All AI responses marked appropriately
  • Modular handlers - Clean separation of concerns across files
  • Pattern matching - Uses app.on_message_pattern for command routing
  • Type safety - Full pyright compliance with proper typing

Architecture

The sample follows a modular architecture:

  • main.py - Main application with pattern-based message handlers
  • handlers/ - Separate modules for different AI functionality:
    • function_calling.py - Pokemon and weather function implementations
    • memory_management.py - Stateful conversation handling
    • citations.py - Citation demo functionality
    • plugins.py - Custom AI plugin implementation
    • feedback_management.py - Message feedback collection and storage

This structure mirrors the TypeScript AI test implementation for consistency across language implementations.