Answer Questions Using Your Own Data

This is a set of 3 Semantic Kernel Python concept samples demonstrating Retrieval-Augmented Generation via Azure OpenAI's On Your Data API: a baseline chat-with-data sample (azure_chat_gpt_with_data_api.py), a variant that combines it with function calling (azure_chat_gpt_with_data_api_function_calling.py), and a variant using vector search specifically (azure_chat_gpt_with_data_api_vector_search.py).

Use these when you want Azure OpenAI to handle the RAG retrieval step itself via its native On Your Data API, rather than building and managing a separate vector store and retrieval pipeline yourself. The function-calling variant is useful when the agent also needs to invoke tools alongside grounded data retrieval; the vector-search variant is the fit when you specifically want vector-based (as opposed to keyword or hybrid) retrieval against your data source.

Not the right samples if you're not using Azure OpenAI, or if you need full control over the retrieval pipeline (custom chunking, reranking, hybrid search tuning) beyond what the managed On Your Data API exposes - a custom vector-database-backed RAG pipeline would be needed for that level of control.

Inputs: an Azure OpenAI deployment configured with On Your Data (pointing at your data source), plus user chat messages.

Outputs: chat completions grounded in the connected data source, optionally combined with function-calling tool invocations or restricted to vector-search retrieval specifically.

Microsoft Semantic Kernel (Python), Azure OpenAI On Your Data API.

Developers on Azure OpenAI who want managed RAG via On Your Data rather than building a custom retrieval pipeline, and need the specific variant matching whether they also need function calling or vector-search-specific retrieval.