Build an AI Agent with MongoDB and n8n

Open the credential configuration dialog.

Click +, and then click New credential.

Add credentials for your model provider.

1. From the drop-down menu, select the credentials to configure.

   For example, select OpenAI and click Continue.

2. To see which parameters to enter for your model provider, check the Credentials Library.

   For example, for OpenAI, enter the API Key, the name of the organization, and the base URL.

3. Click Save. n8n adds the credential to your personal space.

Add credentials for MongoDB.

Follow the same instructions described in the previous step to add your MongoDB credentials to n8n. Specify the following values:

Create a new n8n workflow.

Click +, and then click New workflow.

Add a Manual Trigger node.

1. Click Add first step to open the node menu.

2. Search for Manual Trigger, and then select it to add the node to the canvas.

Add a MongoDB node.

1. Select the + connector on the Manual Trigger node.

2. Search for MongoDB, and then select the MongoDB node.

Configure the MongoDB node.

Select the MongoDB credentials that you configured, and then specify the following values:

For Index Definition, paste the following index definition. This definition indexes the plot_embedding field as the vector type with 1536 dimensions:

{  "fields": [    {      "type": "vector",      "path": "plot_embedding",      "numDimensions": 1536,      "similarity": "dotProduct"    }  ]}

Run the node to create the index.

Click Execute step. When the node runs successfully, the output panel displays the name of the index.

MongoDB builds the index in the background. You can use the index in your workflows after the build completes.

Create a new n8n workflow.

Open n8n where you are hosting it, and then click +. Then, click New workflow.

Add a Chat Trigger node.

All workflows in n8n start with a trigger node. For this workflow, you use a chat trigger node so that the AI agent workflow is triggered by a user message.

1. Select Add first step or press Tab to open the node menu.

2. Search for Chat Trigger. n8n displays a list of nodes that match the search.

3. Select Chat Trigger to add the node to the canvas.

4. Click Back to canvas in the upper-left corner of your screen. You can skip testing this node at this stage. Click Save.

Add an AI Agent node.

In the AI Agent node, you configure the chat model, memory, and tools. These are required components of the AI Agent node where you add additional nodes and sub-nodes.

1. Select the + connector on the trigger node.

2. Search for the AI Agent node and add it to the workflow. The editing view of the AI agent displays.

3. Click Back to canvas in the upper-left corner of your screen. You can skip testing this node at this stage. Click Save.

Configure the chat model.

Configure the chat model to attach the node to process incoming prompts:

1. Click the + connector next to the Chat Model label on the AI Agent node. The search panel appears, filtered on Language Models.

2. Select OpenAI Chat Model from the list, or your preferred LLM provider.

3. Click Credential to connect with.

4. After selecting your credentials, choose a model from the drop-down menu (for example, gpt-4o).

5. Click Back to canvas in the upper-left corner of your screen. You can skip testing this node at this stage. Click Save.

6. Click the AI Agent icon at the top of the dialog box to return to the AI Agent node configuration.

Add the MongoDB Chat Memory sub-node.

Configure the MongoDB Chat Memory sub-node to enable persistent conversation context.

1. Click the + connector next to the Memory label on the AI Agent node.

2. From the side panel, select MongoDB Chat Memory.

3. Click Select credential, and then select your MongoDB credentials.

4. Use the default configurations for the MongoDB Chat Memory node. By default, n8n creates a collection called n8n_chat_histories in the specified database.

5. Click Back to canvas in the upper-left corner of your screen. You can skip testing this node at this stage. Click Save.

6. Click the AI Agent icon to return to the AI Agent node configuration.

Add the MongoDB Atlas Vector Store node as a tool.

Configure the MongoDB Atlas Vector Store node to enable document retrieval.

1. Click the + connector next to the Tool label on the AI Agent node.

2. In the search panel, select MongoDB Atlas Vector Store.

3. Select the MongoDB credentials that you configured, and then specify the following values:

   Setting	Value
   Operation Mode	Retrieve Documents (As Tool for AI Agent) To learn more about this mode, see Retrieve Documents (As Tool for AI Agent).
   Description	"Search through documents about movies to find relevant information"
   MongoDB Collection	embedded_movies
   Embedding	plot_embedding
   Metadata Field	plot
   Vector Index Name	vector_index
   Limit	4
   Include Metadata	Toggle to Off IMPORTANT: If you don't toggle this setting to Off, the AI agent might not run as the documents might be too large to load into the context window.
   Rerank Results	Toggle to Off

Configure the embedding model.

Configure the embedding model for the MongoDB Atlas Vector Store node:

1. Click the + connector next to the Embedding label on the MongoDB Atlas Vector Store node.

2. From the side panel, select Embeddings OpenAI.

3. Select your OpenAI credentials, and then specify the text-embedding-ada-002 model.

4. Click Back to canvas in the upper-left corner of your screen. Click Save.

Test the workflow.

Now that you've configured the AI Agent node, you can test the complete workflow. After you send a message, the workflow runs in real-time on the canvas.

1. Click the Open chat button near the bottom of the canvas.

2. Test the memory functionality by running a few initial prompts. For example:

   Hi, my name is Mongo

   View output

   What is my name?

   View output

3. Test the vector search and RAG functionality by running the following prompt.

   Recommend me a few movies about time travel

   View output

Save the workflow.

1. Click the Save button in the top right of the editor window.

2. Give your workflow a descriptive name. For example, name it "AI Agent with MongoDB".

You can return to this workflow at any time.