Resources · n8n integration

n8n Azure AI Search Vector Store nodeConfigure Azure AI Search Vector Store in n8n.

Azure AI Search stores your embeddings, and the n8n Azure AI Search Vector Store node is how n8n talks to it. Two head parameters drive it: an Operation Mode with 5 choices, from inserting documents to exposing the index as an agent tool, and the Index Name n8n creates when it is missing.

Try the Azure AI Search Vector Store node on n8nSee the 2 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Azure AI Search Vector Store node actually do?

Azure AI Search (formerly Azure Cognitive Search) is a cloud search service with vector search capabilities for RAG and semantic search. This node stores, retrieves and queries vector embeddings, which are numeric representations of your text, alongside the content and metadata they came from. It runs like a regular action node, once per incoming item, and it has no sub-node port of its own.

Scenario one: a searchable knowledge base. Documents arrive in a workflow, an embeddings model turns them into vectors, and the node in insert mode uploads them into an index. Batch Size controls how many documents go up per batch to Azure AI Search, which matters once your source files get long. Nothing else in the workflow needs to know the index schema, because the node creates the index if it is missing.

Scenario two: answering questions inside an agent. Set Operation Mode to retrieve-as-tool and the index becomes a tool that an AI Agent can call on its own. The Name and Description fields are what the model reads to decide whether the tool fits the question, so a vague description is the fastest way to get an agent that never searches. Pair it with a chat model from OpenAI, Anthropic or Google Gemini, whichever provider you already have a key for.

Scenario three: a one-off ranked lookup. In load mode the node takes a Prompt, ranks documents by similarity and hands the results to the next step, which can be a Slack message or a row appended in Google Sheets. Include Metadata decides whether the source fields travel with the text or not.

When to pick something else: if the documents never change and you only ever look up a handful of them, a plain lookup in a table is cheaper to run than an index. The vector store earns its place once similarity matters more than exact matching, or once an agent needs to search without you writing the query.

The known limits are worth reading before you commit. The index is case-sensitive on its name. Your embeddings model output has to match the vector dimensions of the index, otherwise inserts fail on a dimension mismatch. Write operations need the admin API key, a query key only reads. And the node covers what its 2 head parameters expose, no more: index administration, schema edits and service provisioning stay in the Azure portal.

Connect

What do you need to connect before the node runs?

  1. 01

    Create the search service in Azure

    The node has three prerequisites on the Azure side. An Azure subscription, an Azure AI Search service running inside it, and API key authentication configured on that service. Once the service exists it exposes an endpoint in the form https://your-service.search.windows.net, and that URL is the value n8n asks for. Nothing here is created by n8n, so start in the Azure portal and come back with the endpoint in hand.

  2. 02

    Pick the right API key

    Azure AI Search issues two kinds of keys and the choice is not cosmetic. An admin key covers write operations, so it is the one you need for insert and update. A query key is read-only, enough for load, retrieve and retrieve-as-tool. If a workflow only ever searches, give it the query key and keep the admin key out of the instance. Document insert failures that look like a schema problem often come down to a key without write permission.

  3. 03

    Save the credential once in n8n

    In n8n, the Azure AI Search credential is created once from the Credentials menu with the endpoint and the key, then reused by every workflow that touches the service. That applies on n8n Cloud and on a self-hosted instance alike. The node has no Authentication selector in its panel: the credential is the only place authentication is set, and switching from a query key to an admin key later means editing the credential, not the node.

Parameters

Which parameters do you set, and when?

The Azure AI Search Vector Store node has 2 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Operation Mode

mode

What you see in n8n

Notes & use cases

Set this selector first: it decides which fields the panel shows.

Key parameters

  • Operation Mode: insert (Insert Documents) writes documents, load (Get Many) fetches ranked documents for a query, update (Update Documents) updates them by ID, retrieve exposes the store to a chain or tool, retrieve-as-tool to an AI agent.
  • Name and Description: required in tool mode, both read by the LLM.
  • Prompt: the search prompt behind similarity ranking, often {{ $json.question }}.
  • Limit, Include Metadata, Rerank Results: how many top results return, whether metadata travels with them, whether they are reranked.
  • ID: the ID of an embedding entry, required when updating.
Use cases
one workflow inserts support articles nightly, a second answers from them in tool mode.
02

Index Name

indexName

What you see in n8n

Notes & use cases

The name of the Azure AI Search index the node reads from or writes to. It is required in every mode, and n8n creates the index automatically if it does not exist yet, which means a typo does not raise an error: it quietly builds a second, empty index.

Key parameters

  • Index Name: a plain string, case-sensitive, pointing at an index in your service; leave it static for a fixed knowledge base, or drive it from the item with {{ $json.index }} when one workflow serves several collections.
Use cases
separate indexes per language or per customer, selected from the incoming item, so a single insert workflow feeds all of them without branching.
Need help

Need help automating Azure AI Search Vector Store with n8n?

A person reads every message.

FAQ

Questions people ask next

01Is the n8n Azure AI Search Vector Store node included in n8n?
Yes. It ships inside the n8n AI package, @n8n/n8n-nodes-langchain, which comes with n8n itself. There is nothing to install and no extra cost on the n8n side, on n8n Cloud and on a self-hosted instance under the Community Edition and its Sustainable Use license alike. A workflow built in one place behaves the same in the other. What does cost money sits outside n8n: your Azure AI Search service is billed by Azure, and the embeddings model provider bills its own API calls under its own terms. n8n adds nothing to either bill.
02What credentials does the node need?
One credential, for Azure AI Search itself. You authenticate the search service with your endpoint, in the form https://your-service.search.windows.net, and an API key from that service. Which key depends on what the workflow does: an admin key is required for write operations, insert and update, while a query key is read-only and covers search modes. The prerequisites are an Azure subscription, a running Azure AI Search service and API key authentication configured on it. The node itself has no Authentication selector, so the credential is set once from the Credentials menu in n8n and reused everywhere.
03What are the limits of this node?
It exposes 2 head parameters, and that is the whole surface: Operation Mode with its 5 choices, and Index Name. Index administration stays in Azure, so schema design, service tiers and access policies are configured in the portal, not here. Three failure modes come up often. An index name that does not match, since names are case-sensitive and a wrong one silently creates an empty index. A vector dimension mismatch, when the embeddings model output does not match the index vector field. And insert failures caused by a key without write permission or documents that do not match the index schema. The node also has no sub-node port of its own: it runs as a regular action node on each incoming item.
04What do you need to plug in to get started?
Beyond the credential, an embeddings model whose output dimensions match the index, and an index name. Start in insert mode with a small batch of documents and let n8n create the index for you, which sets the vector fields from your embeddings model, an HNSW graph for similarity search with a cosine metric, plus content and metadata fields. Check the results in Azure, then switch a second node to load mode with a Prompt to confirm the ranking looks sane. Only once that works is it worth wiring retrieve-as-tool behind an agent, because a tool that returns nothing is much harder to debug than a query that returns nothing.
05n8n or Make for a vector store workflow?
It depends on where the data has to live. Make is a hosted automation platform with no self-hosting option, billed per operation, so a workflow that pushes thousands of documents through an index has a cost that scales with volume. n8n runs on n8n Cloud or on your own infrastructure, which matters when documents are sensitive and you would rather they never leave your network, especially alongside an Azure service reachable through a private endpoint. Both build visually. The honest split: hosted convenience against control over data and cost model.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations