A BI system requires skills
An executive has to know SQL or wait for a report from an analyst. A simple question about data turns into a task that takes hours or days.
SUB–04 / DATA · Data Analyst · PostgreSQL · Voice and text
Querying PostgreSQL in natural language by text or voice.
A data analyst for a PostgreSQL database. It queries the database in natural language by text or voice: it takes a question, translates it into an SQL query to the permitted views and returns an answer with an explanation of the calculation and a link to the source table. It does not change the data — it only reads. It can run locally within an approved environment or use cloud models; data-handling terms depend on the selected configuration.
01Problem
Why getting an answer about your data is still hard.
Four reasons to use a specialized agent instead of a BI system or cloud AI.
The analyst is a bottleneck
Every question about data goes through the analyst. The queue of requests grows, and teams have to wait for even simple metrics.
Cloud processing requires review
In cloud mode, queries and results are processed in the selected provider's infrastructure. Review the agreement, storage, access, and data-use terms before connecting it.
Typing is not always convenient
In a meeting, on the road or on the production floor, typing a query is inconvenient. Voice input solves this problem.
The Data Analyst accepts a question by voice or text, builds the SQL query, and links the answer to the source table. In a local configuration with no external integrations, processing remains within the approved environment.
02Process
From a voice question to an answer with data.
Four steps — from speech recognition to an SQL query to PostgreSQL.
Voice or text
An executive asks a question by voice or in text. The agent recognizes the speech and extracts the business meaning of the request.
Access check
The agent identifies the user, their role and the data area they have access to.
SQL query to PostgreSQL
The agent builds an SQL query to the permitted views of the database. Writing and modifying data is prohibited.
Answer with a source
The agent returns the result, explains how it was calculated and points to the source table for verification.
03Advantages
Why this subagent is worth considering.
The design decisions built into the foundation determine the nature of the result and the scope of application.
01
Voice queries
An executive asks a question by voice — the agent recognizes the speech, builds an SQL query and returns an answer. No access to a BI system is needed.
02
Working inside a closed perimeter
In a local configuration with no external integrations, data, queries, and results are processed within the approved company environment.
03
Read-only access
The agent builds SQL queries only to the permitted PostgreSQL views. Writing and modifying data is prohibited at the architecture level.
04
Local or cloud models
The engine supports local and cloud models. The user selects the configuration, and data flows and access terms are defined before launch.
04Audience
Who gets the most value from the subagent.
Each role gets a concrete result that fits into its own workflow.
Executives
Operational metrics in response to a voice request, without going to analysts or BI systems.
Analysts
Fast verification of data and metrics without writing SQL queries.
Finance teams
Revenue, margin and receivables metrics — by voice or in text.
Production
Real-time operational data on the warehouse, production and deliveries.
05Scenarios
Where the agent saves hours of manual work.
Typical situations: from a voice query in a meeting to financial monitoring.
A voice query in a meeting
An executive asks: "What was the revenue for the last quarter by region?"
What is generated
- Speech recognition
- An SQL query to PostgreSQL
- An answer with the figure and the source table
What it gives
- An answer without waiting for a manually prepared report
- No laptop or BI system needed
An analyst checks a metric
An analyst needs to quickly verify a metric without writing SQL.
What is generated
- A natural-language text query
- Data from a permitted view
- An explanation of the calculation and the source
What it gives
- Faster verification without writing SQL manually
- No need to know the database schema
Financial monitoring
A CFO checks the receivables.
What is generated
- A voice or text query
- Data from a financial view
- A link to the table for audit
What it gives
- Monitoring without waiting for a report
- Every figure is linked to a source
Boundaries
What the human checks.
The answer depends on the quality of the data model and the dictionary of metrics. The agent has no write access and does not bypass corporate access policies. Access to sensitive data is restricted through a role-based access model.
What the agent does not do
- Does not modify data in the database — read-only access only
- Does not bypass corporate access policies
- Does not work with data the user has no rights to
- Does not invent data — only real results of an SQL query
06Questions
Short answers to frequent questions.
Direct answers for quick reading. Expand the question you need or open all at once.
Can the agent change data in the database?
No. The agent has read-only access to the permitted PostgreSQL views only. Writing and modifying data is prohibited at the architecture level.
Is the data sent to an external service?
In local mode — no. Data, queries and results stay inside the company perimeter. Cloud mode — only by the user's explicit choice.
Can I ask questions by voice?
Yes. The agent recognizes speech (locally via Whisper or a cloud STT) and builds an SQL query from the meaning of the question.
Which databases are supported?
PostgreSQL. The agent works with permitted views, not with tables directly.
Do I need to know SQL?
No. The question is asked in natural language — in text or by voice. The agent builds the SQL query itself.
How is access restricted?
Separate views with access rights for different roles are supported. Access to sensitive data is restricted through a role-based access model.
First step
Choose a ready-made solution or describe your process.
Pick one of the twelve agents and subagents, or fill in the questionnaire for custom development.
