mongodb_atlas_vector_search_setup.md

Author : Georgios Ioannou

Copyright © 2024 by Georgios Ioannou

Setting Up MongoDB with Atlas Vector Search

Initial Setup

1. Create and log in to your MongoDB account at mongodb.com

2. Create Organization and Project

   • Create a new MongoDB organization
   • Name your organization as desired
   • Create a new MongoDB project
   • Name your project as desired

3. Create Cluster

   • Select M0 (FREE) tier
   • Wait 1-3 minutes for cluster creation

Security Setup

4. Create Database User

   • Create username and password
   • Important: Save these credentials securely

5. Configure Network Access

   • Navigate on the left menu to Security -> Network Access
   • Add IP Address: 0.0.0.0/0 (Allows access from anywhere)

Database Creation

6. Set Up Database and Collection
   • Click "Browse Collections"
   • Click "Add My Own Data"
   • Create database (e.g., "txts")
   • Create collection (e.g., "txts_collection")
   • Click "Create"

Vector Search Configuration

7. Create Vector Search Index
   • Click "Search Indexes" (This will bring you to "Atlas Search" tab)
   • Click "Create Search Index"
   • Select "JSON Editor" under "Atlas Vector Search"
   • Click "Next"
   • Configure index settings:

     {
       "name": "vector_index", // Adjust it to your desired name.
       "type": "vector",
       "path": "embedding", // Adjust it to your index/embedding column.
       "numDimensions": 384,  // Adjust based on your model. (e.g., 384 for all-MiniLM-l6-v2)
       "similarity": "euclidean" // Adjust it to your desired similarity function.
     }
     

   • Optional: Add filter fields if needed:

     {
       "type": "filter",
       "path": "source" // Adjust it to the column you would like to filter with.
     }
     

   • Assign your index to the collection name created in step 6 using the drop down menu on the left.
   • Click "Next"
   • Click "Create Search Index"
   • Wait for status to change from "Pending" to "Active"

Note: While vector search index creation can be done programmatically, the GUI method described above is recommended for simplicity.

Connection String

8. Get Your MongoDB URI
   • Click "Clusters" under DATABASE on the left menu
   • Click "Connect" next to name of your cluster
   • Select "Drivers"
   • Copy the connection/MongoURI string at the very end of the modal that just opened
   • Replace <password> with your database user password