Agentic data analysis for PostgreSQL, MySQL, Oracle & SQL Server

Ask your data anything.

Your database holds the answers. QueryAI uncovers them with natural-language questions, context-aware SQL, and clear analytical insights.

See how it works

No data movement. QueryAI reads your schema, runs read-only queries, and writes nothing.

Which products generated the most revenue this quarter?

Try another question

Agent activitymax 2 queries

  1. Question understood

  2. retrieve_context · hybrid schema search

  3. run_sql · validated read-only

  4. Composing the answer

SQL · generated
1SELECT
2 p.name AS product,
3 SUM(o.total) AS revenue,
4 COUNT(*) AS orders
5FROM orders o
6JOIN products p ON p.id = o.product_id
7WHERE o.created_at >= DATE_SUB(CURDATE(), INTERVAL 3 MONTH)
8GROUP BY p.name
9ORDER BY revenue DESC
10LIMIT 5;

$4,820

top product revenue

Top 5 of 24 products by quarterly revenue
Revenue per productValue

Meridian Desk Lamp

$4,820

96 orders

Fieldline Backpack

$3,690

74 orders

Halden Mug Set

$3,156

118 orders

Vantage Desk Mat

$2,480

63 orders

Orla Ceramic Vase

$1,931

41 orders

Top 5 of 24 products by quarterly revenue

Starting demonstration

Sample workspace with demonstration data. Nothing here is executing, connect your own database inside the app.

01 · How it works

From connection to answer in three steps

No dashboards to configure, no SQL to hand-write. QueryAI sits on top of the database you already have.

  1. 01

    Connect

    Point QueryAI at a supported database. The schema (tables, columns, foreign keys) is extracted and indexed for retrieval.

    database_typepostgresql
    hostdb.internal.acme.io
    port5432

    Credentials encrypted (Fernet) before they are stored

  2. 02

    Ask

    Describe the analysis in plain English. The agent retrieves the most relevant tables from your indexed schema before it touches any query.

    How did churn move after the March pricing change?

    Schema context retrieved first, so SQL fits your tables

  3. 03

    Discover

    Inspect the generated SQL, the returned rows, and a written analytical answer that states what the data actually shows.

    SELECT plan, SUM(total)

    FROM v2_mrr_by_plan GROUP BY plan

    Every answer links back to the exact query and rows

Supported engines: PostgreSQL, MySQL, Oracle, and Microsoft SQL Server.

02 · Under the hood

An agent that investigates before it answers

Each question runs through the same honest pipeline: ground the question in your real schema, query carefully, then explain what came back.

Question

Plain English

Schema retrieval

Dense + BM25, top 10

SQL execution

Validated read-only

Results

Up to 50 rows per run

Explanation

The numbers, in words

The agent is bounded: at most two schema retrievals and two SQL queries per question, so it investigates with intent instead of thrashing.

03 · Features

Built for questions, not dashboards

Everything on this page maps to shipped behavior: hybrid schema retrieval, bounded agent runs, read-only execution, conversation memory.

Natural-language exploration

Ask in the words you'd use with a colleague. Follow-ups keep the thread, so refinement doesn't mean starting over.

Top customers by spend this month?

The top 5 account for 62% of monthly revenue. Alice Johnson leads at $15,420.

Ask a follow-up: “Which of them ordered in the last 7 days?”

Context-aware schema retrieval

Your schema is indexed and searched with hybrid dense + sparse retrieval, filtered to your account and database.

T:orders | cols:8 | J:customers(customer_id)

T:products | cols:6 | J:orders(product_id)

T:customers | cols:5

tenant: user__dbdense + BM25top 10

AI-assisted, read-only SQL

Queries are generated for you, validated as read-only, and executed with results capped before they reach the answer.

run_sql · validatedwrite blocked

DROP TABLE orders

returns: rejected before execution

SELECT total FROM orders ✓

Analytical answers

Not just rows: a written answer that names the metric, the comparison, and the conclusion you can act on.

“August orders split 44% repeat vs 56% new; repeat buyers average $61 per order vs $48 for first-time customers.”

metric · comparison · conclusion

Conversation history

Every conversation and turn is stored, so yesterday's investigation is still available today.

  1. Revenue Analysis4 turns
  2. Churn after pricing change2 turns
  3. Top regions by margin6 turns

Preferences that stick

The agent remembers how you like answers delivered and applies it conversation after conversation.

Prefers EURWeekly summary cadenceTimezone UTC+2Runs margin analysis on Mondays

Preferences are remembered per user and per database, and steered into every new answer.

04 · Security

Your database is respected as your database

QueryAI is designed around one rule: it can look, but it cannot touch. These are current product behaviors, not roadmap promises.

  • Read-only by construction

    Every SQL statement is validated before execution. Only reads pass; the agent cannot insert, update, or delete anything in your database.

  • Credentials encrypted at rest

    Connection credentials are encrypted before they are stored and decrypted only in memory, at the moment a query runs.

  • Tenant-scoped context

    Schema retrieval is filtered strictly to your user and your database. One tenant never sees another tenant's tables.

  • Bounded, observable runs

    Each answer is limited to two schema lookups and two SQL queries. You see the exact queries and rows behind every conclusion.

Your data has answers.
Start asking.

Sign up, connect one database, and ask your first real question in plain English.

Back to the demo

Works with PostgreSQL, MySQL, Oracle, and SQL Server