Connect Copilot Studio to Azure AI Search: Build a Knowledge-Powered Agent

Copilot Studio agents are only as smart as the knowledge they can reach. By connecting to an Azure AI Search index, you can ground your agent in your own unstructured content — product manuals, policy documents, PDFs, and more. This integration uses retrieval-augmented generation (RAG) to pull the most relevant pieces of text from your documents and feed them into the generative model.

In this guide you'll set up the full pipeline: an Azure AI Search index backed by Blob Storage, an embedding model deployed through Azure OpenAI, and a Copilot Studio agent that queries the index to answer user questions.

When you're done, your agent will be able to answer questions like "What is the return policy for electronic accessories?" directly from your company's support PDFs — no manual re‑typing required.

Scenario: Contoso Support Knowledge Base

Contoso Electronics stores product manuals and warranty policies as PDFs in an Azure Blob container. Their support team wants a Copilot that can read those documents and give accurate answers without copying text into a custom knowledge store.

The plan:

• Upload manuals to a dedicated Blob container.
• Create an Azure AI Search index that vectors the text using an embedding model.
• Connect that index to a new Copilot Studio agent as a knowledge source.

All resources live in the same Azure region and resource group to keep latency low and management clean.

Step 1 – Create the Required Azure Resources

You need three services before you can touch Copilot Studio.

1.1 Azure AI Search

1. In the Azure portal, search for AI Search and start the creation wizard.
2. Use a Standard tier (or Free for testing) and place it in your chosen resource group. Free only works for very small datasets (up to 50 MB of text).
3. Give the service a globally unique name, for example contoso‑kb‑search.

1.2 Azure Storage (Blob)

1. Create a new Storage account with the same subscription and resource group as the Search service.
2. Choose Azure Blob Storage as the primary service.
3. Inside the storage account, add a container named manuals.

1.3 Azure OpenAI

1. Create an Azure OpenAI resource (Standard S0 tier).
2. After deployment, open the Azure AI Foundry portal (the link inside the resource).
3. Go to Deployments and deploy an embedding model. text-embedding-3-large gives the most accurate vectors, but text-embedding-ada-002 is cheaper and enough for many use cases.
4. Make note of the deployment name; you'll need it later.

Step 2 – Upload Documents to Blob Storage

In the new storage account, open the manuals container and upload your PDF files. Copilot Studio (via the Azure AI Search indexer) can process PDFs, Word documents, and plain text files. For this scenario, upload at least one product manual.

Once the files appear, you can move to indexing.

Step 3 – Deploy an Embeddings Model

Inside Azure AI Foundry, you already deployed an embedding model. If you haven't done that yet, return to the Deployments page and follow the wizard. The model converts every chunk of text into a numerical vector that the search engine can compare against user queries.

Embedding models do not support images; only text inside your documents is vectorized.

Step 4 – Import and Vectorize Data in Azure AI Search

This is where the magic happens. Azure AI Search can scan your Blob container, chunk the documents, call your embedding model, and write vectors into a search index — all in one indexer job.

1. On your Azure AI Search service overview page, click Import and vectorize data.
2. Choose Azure Blob Storage and the RAG scenario.
3. Select your storage account and the manuals container.
4. For the vectorization options, pick Azure OpenAI and fill in:
   • Subscription / resource group
   • The Azure OpenAI service you created
   • The deployment name of your embeddings model
5. Under Advanced settings → Schedule choose Once (or set a recurring schedule if documents update frequently).
6. Review and create the indexer.

The indexer will create a new index automatically and start processing. While it runs, you can monitor progress under Indexers in the Search service.

Step 5 – Verify the Index and Get Connection Details

After the indexer finishes with Success, test it:

1. Go to Indexes and open the newly created index.
2. Use the built-in search bar to type a question like "What is the warranty for a laptop?".
3. The index should return matching chunks from your documents.

Now grab the credentials needed by Copilot Studio:

• Endpoint URL – found on the Overview page of the Azure AI Search service.
• Admin Key – found under Keys (use the primary admin key).

Step 6 – Connect Copilot Studio to Azure AI Search

1. Create a new agent in Copilot Studio (or open an existing one).
2. Go to the Knowledge tab and click Add knowledge.
3. Choose Azure AI Search.
4. Create a new connection:
   • Authentication: Access Key
   • Endpoint URL: paste from the Search overview page
   • Admin Key: paste the primary key
5. After the connection is established, select the index you created (it appears in the dropdown).
6. Click Add to agent.

The agent will now list Azure AI Search as a knowledge source with status Ready.

Step 7 – Test Your Copilot

In the Copilot Studio test pane, ask a question that your documents can answer. For example:

• "How do I reset the Contoso Smart Thermostat?"
• "What is the return period for accessories?"

The agent should pull relevant content from the index and produce an answer based on that information, not on its own training data.

Additional Considerations

Security & delegation

• Prefer managed identity over access keys. Assign the Search Index Data Reader role to the managed identity used by Copilot Studio (currently requires the Power Platform identity to be granted permissions on the Search resource).
• If you need multi‑tenant isolation, use separate indexes per tenant.

Performance and scaling

• The Free tier of Azure AI Search is limited to 3 indexes and 10K documents. Upgrade to Standard as soon as your content grows.
• Indexer scheduling — set it to Once for static documents, or a recurring schedule for frequently updated content.
• Monitor indexer failures via the Indexers tab; common errors include token‑limit exceeded (files too long) or unreachable embedding models.

Troubleshooting common mistakes

Indexer fails after "Import and vectorize data" → Check the error message. Most often it's a token limit in a single document. Reduce the document size or increase the maxStringLength in the indexer's settings (JSON).

Azure AI Search index appears empty → Verify that the embedding model is deployed and accessible from the Search region. Also ensure the Blob container path is correct.

Copilot ignores the search results → Make sure the agent's instructions emphasize using the knowledge source. Add a line like: "Only answer from the provided knowledge."

Final Recommendation

Azure AI Search is one of the most powerful knowledge sources available for Copilot Studio today. It gives you full control over the vectorization, search quality, and document lifecycle. Start with a small set of clean PDF files, test thoroughly, and gradually expand the index.

For a production environment, plan the layout of your indexes — one per topic, department, or product line — and always use managed identity authentication.

References

• Original setup guide by Matthew Devaney: Copilot Studio: Azure AI Search Complete Setup Guide
• Azure AI Search documentation (Microsoft Learn): Chunking and vectorization in Azure AI Search (Add official URL after verifying)
• Copilot Studio knowledge sources: Add knowledge to Copilot Studio (Add official URL after verifying)