OP_DLCTtext2sql.aceanalytics.dev

Ask in English.
Get production SQL.

Dialect routes your question through nine specialized agents - intent, context, schema, planner, writer, validator - before any LLM writes a line of SQL.

9 agents
purpose-built
Schema-valid
every output
Self-correcting
repair loop
dialect.ace / query_session_0417
● CONNECTED · banking_core
QUERY"Show average deposit balance by market for Q4"
Intent
write_sql · aggregation · time_filter
12ms
Context
3 tables ranked · 12 fields selected
45ms
Planner
GROUP BY market · AVG(balance)
28ms
Validator
schema-valid · joins OK
19ms
GENERATED SQL
-- validated · schema-safe
SELECT d.market,
       AVG(b.balance) AS avg_deposit
FROM deposits b
JOIN dim_market d ON b.market_id = d.id
WHERE b.period_end BETWEEN '2026-10-01'
      AND '2026-12-31'
GROUP BY d.market
ORDER BY avg_deposit DESC;
OP_DLCT_02·The 9-agent pipeline

Nine agents.
One question.

01
Intent
Classifies request type and required operators
02
Context
Retrieves candidate tables and fields from the semantic layer
03
Reranker
Scores candidates, applies dimensional modeling preferences
04
Glossary
Maps business terms (ROE, NIM, market) to columns
05
Planner
Decides joins, filters, grouping, and order
06
Writer
Emits SQL using the planner's structure
07
Validator
Schema, join, and type checks. Fails closed.
08
Repair
On failure, revises and retries - up to n times
09
Explainer
Produces plan-of-reasoning and cell-level provenance
01

Banking semantic layer

Domain-specific table descriptions, column relationships, and business glossary baked into every plan.

02

Context reranking

Candidate tables and fields are scored by relevance, with dimensional modeling preferences applied.

03

Self-correcting SQL

Validation failures trigger an automatic repair loop. Schema errors are fixed before you see them.

04

Explainable plans

Every query comes with a step-by-step breakdown of what it does and why each join was chosen.

OP_DLCT_03·Flow

Ask. Enrich. Execute.

01

Ask in plain English

Type "Show average deposit balance by market for Q4". Intent is classified instantly.

02

Multi-agent pipeline

Nine agents collaborate to fetch context, rank fields, plan structure, and validate SQL.

03

Execute & visualize

Run against live banking data. Explore results in interactive tables and charts.

The best prompt
is richer context.

Dialect ships with the banking semantic layer. Plug in your schema, get answers.