How to Run an AI MongoDB NoSQL Agent with OE Runtime

Query and analyse your MongoDB collections with natural language. OE Runtime connects to any MongoDB instance, explores your schema, surfaces data quality issues, and suggests indexes — all without writing a single aggregation pipeline.

The Agent YAML

Create a file called agent.yaml inside a nosql-cache/ directory:

name: NoSQL Data Agent
description: Read and write MongoDB documents
instructions: |
  You are a data agent with access to a MongoDB database.
  Query, insert, and analyze documents. Always confirm before write operations.
  Complete all steps fully before writing your report.
steps:
  - name: Query Collection
    content: |
      List all available collections in the database.
      From the first collection, fetch the 10 most recently created documents.
      Note the schema structure — list all field names and their value types.
  - name: Analyze Documents
    content: |
      From the retrieved documents:
      - Identify the most common field values (top 3 values for any categorical fields)
      - Find any documents with missing or null fields
      - Calculate average numeric values if any numeric fields exist
  - name: Report
    content: |
      Produce a data summary:
      - Collections found in the database
      - Schema of the queried collection
      - Key statistics from the 10 documents
      - Documents with data quality issues (nulls, missing fields)
      - Suggested indexes based on the field patterns observed
connectors:
  - connection_name: My MongoDB
    connection_type: mongodb

The Config File

Create oe-config.json in the same directory:

{
  "llm": {
    "provider": "openai",
    "model": "gpt-4o",
    "apiKey": "YOUR_OPENAI_API_KEY"
  },
  "server": {
    "enabled": false,
    "port": 3333,
    "apiKey": "your-secret-api-key"
  },
  "connectors": [
    {
      "connection_name": "My MongoDB",
      "connection_type": "mongodb",
      "uri": "mongodb://localhost:27017",
      "database": "mydb"
    }
  ]
}

Replace the URI with your MongoDB connection string. For MongoDB Atlas use the connection string from your cluster's Connect dialog. The agent works with MongoDB 4.x and later, including Atlas free tier clusters.

Download OE Runtime

Run the Agent

Windows

# Run the agent
oe-runtime-win.exe nosql-cache/agent.yaml --config nosql-cache/oe-config.json

macOS

# Make executable (first time only)
chmod +x oe-runtime-macos

# Run the agent
oe-runtime-macos nosql-cache/agent.yaml --config nosql-cache/oe-config.json

Linux

# Make executable (first time only)
chmod +x oe-runtime-linux

# Run the agent
oe-runtime-linux nosql-cache/agent.yaml --config nosql-cache/oe-config.json

Running as an API Server

Add --serve to start OE Runtime as an HTTP server on port 3333:

oe-runtime-win.exe --serve --config nosql-cache/oe-config.json

Then send a request:

curl -X POST http://localhost:3333/run \
  -H "Content-Type: application/json" \
  -d '{"yaml": "nosql-cache/agent.yaml", "params": {}}'

Or import the Postman collection (download above) to test all endpoints visually.

Use Cases

• Schema discovery and documentation — Automatically explore a MongoDB database and document the schema of every collection, including field names, types, and example values. This is invaluable when onboarding to an undocumented legacy database or preparing for a data migration project.
• Data quality auditing — Identify documents with null, missing, or inconsistent fields across large collections without writing complex aggregation queries. The agent surfaces data quality issues and counts affected documents so your engineering team knows where to focus cleanup efforts.
• Self-service analytics — Let non-technical stakeholders query MongoDB collections in plain English without needing to understand aggregation pipelines or JSON query syntax. The agent translates natural language questions into MongoDB queries and returns formatted results.
• Index optimisation — Analyse query patterns across your collections and recommend compound indexes that would improve read performance. The agent identifies fields that are frequently filtered or sorted but lack indexes, and generates the index creation command ready to run.
• User behaviour analysis — Query a MongoDB collection storing user events, session data, or clickstream records to answer product analytics questions. The agent can group events by user, calculate engagement metrics, and surface behavioural patterns without a dedicated analytics database.
• Content management — Query and update content stored in MongoDB that powers a headless CMS or blog platform. The agent can retrieve recent posts, check for drafts that have not been published, and surface content with missing required fields.
• IoT data analysis — Analyse time-series sensor readings stored in MongoDB collections to identify trends, anomalies, and threshold breaches over a given time period. MongoDB's flexible document model makes it a popular choice for IoT data, and the agent can work with nested sensor reading documents directly.
• E-commerce order analysis — Query order, product, and customer collections in a MongoDB-backed e-commerce platform to answer business questions about revenue, conversion, and fulfilment. The agent surfaces the top-selling products, average order value, and orders stuck in specific status states without any dashboard configuration.