A query agent is an AI component that translates natural language into executable queries across SQL databases, NoSQL databases, and REST APIs. Instead of writing SQL, MongoDB aggregation pipelines, or Elasticsearch DSL by hand, you describe what data you need in plain English and the agent writes, optimizes, and runs the query against the correct data source. Query agents are the foundational layer of agentic BI: every other agent (Dashboard, Widget, Recommendation) depends on the query agent to retrieve data before it can do anything else.
TL;DR
- A query agent converts natural language to executable queries across 55+ data sources, including SQL, NoSQL, and REST APIs.
- It is the foundational agent in agentic BI. Dashboard, Widget, and Recommendation agents all depend on it to retrieve data before they can act.
- Key differentiator: queries NoSQL databases like MongoDB and Elasticsearch natively, using their own query languages, without requiring SQL translation or a warehouse layer.
- Supports cross-source joins: query a MongoDB collection and a PostgreSQL table, then join the results without ETL or data movement.
- Works with Private AI, so the LLM runs on-prem and no data leaves your environment.
- Available in-platform, via MCP (Model Context Protocol), Slack/Teams, and as an embedded component in SaaS products.
- Part of Knowi’s multi-agent orchestration system. Learn more at AgenticBI.com.
Table of Contents
Why Query Agents Matter
Most BI tools assume your data is already in a warehouse, modeled, and ready to query. That assumption breaks down in organizations running MongoDB for application data, Elasticsearch for logs, PostgreSQL for transactions, and a handful of REST APIs for third-party data. Someone still has to write the queries. That someone is usually an engineer or analyst with source-specific expertise.
A query agent removes that bottleneck. A product manager asks “what were the top 10 customers by order volume last quarter?” and the agent identifies the right data source, generates the query in the correct dialect (SQL, MQL, Elasticsearch DSL), executes it, and returns structured results. No ticket to the data team. No waiting for a dbt model.
This is different from a general-purpose chatbot that generates SQL snippets you have to copy, paste, and debug. The query agent is connected to your live data sources, executes the query directly, and passes the results to downstream agents for visualization or analysis.
How Query Agents Work
Natural Language to Query Translation
The query agent parses a natural language request and translates it into the appropriate query language for the target data source. “Show me all orders over $1,000 from the last 30 days” becomes a SQL WHERE clause against a PostgreSQL table, a MongoDB find() with $gte operators against a document collection, or an Elasticsearch bool query with range filters, depending on where the data lives.
The agent reads the data source schema (table structures, collection fields, nested document paths) to generate accurate queries. It understands field names, data types, and relationships without requiring a pre-built semantic model.
Native NoSQL Querying
This is where query agents diverge from every text-to-SQL tool on the market. Most AI query tools only generate SQL. If your data is in MongoDB, they either cannot access it or force you to use a BI Connector that flattens your documents into relational tables, losing the nested structure that made MongoDB useful in the first place.
Knowi’s Query Agent generates native MongoDB aggregation pipelines, native Elasticsearch DSL, and native Cassandra CQL. A query against a MongoDB collection with nested arrays of subdocuments produces a proper $unwind and $group pipeline, not a SQL approximation. This preserves the full expressiveness of each database’s query language.
Cross-Source Joins
A single question often spans multiple data sources. “Show me customer support tickets from Zendesk alongside their purchase history from our PostgreSQL database” requires data from two systems. Traditional BI solves this by ETL-ing everything into a warehouse first. That adds days of pipeline work and ongoing maintenance.
The query agent handles this directly. It queries each source in its native language, retrieves the results, and joins them in memory using Knowi’s Cloud9QL post-processing layer. No warehouse required. No data replication. The join happens at query time.
The Orchestration Layer
The query agent is the first agent invoked in almost every agentic BI workflow. When a user says “create a dashboard showing monthly revenue by region,” the orchestration engine routes the request through multiple agents in sequence:
- Query Agent identifies the data source, writes the query, and retrieves the result set
- Widget Agent maps the fields to chart axes and selects the visualization type
- Dashboard Agent arranges widgets into a layout and configures filters, sharing, and exports
- Recommendation Agent analyzes the data and surfaces anomalies or trends
Without the query agent, none of the downstream agents have data to work with. It is the foundation the entire system depends on.
What Data Sources Does a Query Agent Support?
Knowi’s Query Agent connects to 55+ data sources natively:
- SQL databases: PostgreSQL, MySQL, SQL Server, Oracle, Snowflake, BigQuery, Redshift, ClickHouse
- NoSQL databases: MongoDB, Elasticsearch, Cassandra, DynamoDB, Couchbase, Apache HBase
- APIs: Any REST API endpoint, with support for pagination, authentication, and nested JSON response handling
- Files and feeds: CSV, JSON, XML, Google Sheets
- Cloud platforms: Salesforce, Google Analytics, AWS CloudWatch
Each source is queried in its native language. The agent does not force every source through a SQL abstraction layer. See the full list in the Knowi AI tools documentation.
Query Agent vs. Traditional BI vs. Chatbot Analytics
| Capability | Traditional BI (Tableau, Power BI) | Chatbot Analytics (ChatGPT, Copilot) | Query Agent (Knowi) |
| Query generation | Manual: user writes SQL or uses a visual query builder | Generates SQL snippets, user must copy and execute manually | Generates and executes queries automatically from natural language |
| NoSQL support | Requires BI Connector or warehouse ETL to flatten documents | Generates SQL only; no native MongoDB or Elasticsearch support | Native query generation in MongoDB MQL, Elasticsearch DSL, Cassandra CQL |
| Cross-source joins | Requires all data in one warehouse first | Cannot join across sources | Queries each source natively and joins results at query time without ETL |
| Data source breadth | Primarily SQL warehouses and extracts | Uploaded files or single pre-connected source | 55+ sources: SQL, NoSQL, REST APIs, files, cloud platforms |
| Execution | Connected to live data, runs queries in real time | Returns code, does not execute against your data | Connected to live data, executes queries and returns structured results |
| Multi-agent coordination | Not applicable | Single model, no agent specialization | Feeds results to Dashboard, Widget, Document AI, and Recommendation agents |
| Data privacy | Data in vendor cloud or warehouse | Data sent to LLM provider (OpenAI, Microsoft) | Private AI option: LLM runs on-prem, no data leaves your environment |
Where Query Agents Run
The query agent is accessible through four surfaces, each suited to different workflows:
- In-platform: The AI Assistant panel inside Knowi, where users type questions and the agent queries data, builds charts, and creates dashboards in response.
- MCP Server: Connect Claude, GPT, or any MCP-compatible client to Knowi through the Model Context Protocol. Query your databases from a terminal, IDE, or custom application.
- Messaging: Ask questions from Slack or Microsoft Teams. The query agent retrieves data and returns results or visualizations directly in the channel.
- Embedded: Embed the conversational agent in your SaaS product. Your end users query their own data through natural language, with multi-tenant row-level security enforcing data isolation.
How the Query Agent Feeds the Rest of the System
The query agent does not work in isolation. It is the data retrieval layer for every other agent in the system:
- Dashboard Agent: Receives query results and builds multi-widget dashboard layouts with filters, exports, and sharing.
- Widget Agent: Takes the structured result set and maps fields to chart axes, dimensions, and measures to create individual visualizations.
- Document AI Agent: Uses the query agent to pull structured data alongside unstructured document content for combined analysis.
- Recommendation Agent: Analyzes query results to surface trends, anomalies, and actionable insights.
A single user prompt can invoke up to five agents per request. The orchestration engine manages intent detection, agent sequencing, and shared context so each agent’s output is available to the next.
Frequently Asked Questions
What is a query agent?
A query agent is an AI component that translates natural language questions into executable database queries. It identifies the correct data source, generates the query in the appropriate language (SQL, MongoDB MQL, Elasticsearch DSL), executes it, and returns structured results. Unlike text-to-SQL chatbots, a query agent is connected to your live data and runs the query directly.
How is a query agent different from a text-to-SQL tool?
Text-to-SQL tools only generate SQL and leave execution to you. They also cannot query NoSQL databases natively. A query agent generates queries in multiple languages (SQL, MQL, DSL, CQL), executes them against your connected data sources, and passes the results to other agents for visualization and analysis. It handles the full cycle from question to answer.
Can a query agent query MongoDB without flattening the data?
Yes. Knowi’s Query Agent generates native MongoDB aggregation pipelines that preserve nested document structure, arrays, and subdocuments. It does not require a BI Connector or warehouse that flattens documents into relational tables. This means queries against deeply nested JSON structures return accurate results without losing data fidelity.
What is a cross-source join and how does the query agent handle it?
A cross-source join combines data from two different databases in a single query. For example, joining MongoDB customer records with PostgreSQL transaction data. The query agent queries each source in its native language, retrieves the results, and joins them using Knowi’s Cloud9QL post-processing layer. No ETL pipeline or warehouse is required.
Is my data sent to an external LLM when using a query agent?
Not if you use Private AI. Knowi offers on-prem deployment where the LLM runs entirely within your infrastructure. No data or query content leaves your environment. This is required for organizations handling PHI (HIPAA), financial data, or classified information where data residency is non-negotiable.
How does the query agent fit into the agentic BI orchestration system?
The query agent is the foundational layer. When a user asks a question, the orchestration engine invokes the query agent first to retrieve data. That result set is then passed to the Widget Agent (for visualization), Dashboard Agent (for layout and sharing), or Recommendation Agent (for insights). Up to five agents can coordinate per request through shared context.
What does a query agent cost?
Pricing depends on deployment. AgenticBI.com offers self-serve conversational analytics. Knowi Enterprise includes the agent system with Private AI, embedding, and on-prem deployment. Contact the team for enterprise pricing.