Integrated Embeddings sample app

End-to-end sample for Integrated Embeddings in Azure Cosmos DB for NoSQL. Companion to the announcement blog post.

The app creates a container with an embeddingSource policy, inserts outdoor-product items, lets Azure Cosmos DB generate the embeddings, and runs a vector search.

Prerequisites

• Python 3.10+
• An Azure Cosmos DB for NoSQL account with vector search and change feed mode enabled
• A Microsoft Foundry embedding model deployment (text-embedding-3-small, text-embedding-3-large, or text-embedding-ada-002)
• (RAG agent only) A Microsoft Foundry chat model deployment (e.g., gpt-4o, gpt-4o-mini)
• The Cosmos DB account's managed identity granted Cognitive Services OpenAI User on the Foundry resource
• Your signed-in principal granted Cosmos DB Operator (Azure RBAC) and Cosmos DB Built-in Data Contributor (Cosmos DB RBAC) on the Cosmos DB account

See the blog post and the Integrated Embeddings prerequisites for details.

Setup

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

cp .env.example .env
# edit .env with your account details and Foundry API key

az login

Run

# Step 1: create the database and container with embeddingSource policy
python create_db_and_container.py

# Step 2: insert 100 sample items (embeddings generate in the background)
python insert_sample_data.py

# Step 3: in the Azure portal Data Explorer, run this query until count reaches 100
#   SELECT VALUE COUNT(1) FROM c WHERE IS_DEFINED(c.embedding)

# Step 4: run a vector search
python vector_search.py "I need to stay warm on a cold ski trip"
python vector_search.py "lightweight cookware for backpacking" --top-k 3

# Optional: chat with a RAG agent over the catalog
python rag_agent.py

Files

File	Purpose
config.py	Loads .env and exposes settings to the other scripts.
create_db_and_container.py	Creates the database and container with the vector embedding policy.
insert_sample_data.py	Upserts items from items.json.
vector_search.py	Embeds a query and runs VectorDistance().
rag_agent.py	LangChain RAG agent over the catalog.
items.json	100 sample outdoor-product items.
.env.example	Template for the environment variables. Copy to .env.

Auth

• Azure Cosmos DB calls use Microsoft Entra ID via DefaultAzureCredential.
• The Cosmos DB to Foundry call (Integrated Embeddings itself) uses the Cosmos DB account's managed identity (authType: "Entra" in the policy).
• The query-time call in vector_search.py and rag_agent.py uses an API key (FOUNDRY_API_KEY) to call Foundry for the query embedding (and chat, for the RAG agent).