Natural Language to SQL, Built for Enterprise

Turn business questions into traceable SQL workflows

Text2SQL transforms business questions into executable, auditable, and governable SQL with multi-source access, RAG context, LangGraph orchestration, safety controls, and streaming delivery.

View Architecture

Live Query Console

Natural language question run_01H8Z...

How many orders and how much revenue did each region generate last month? Sort by revenue descending.

RAG Context

orders schemaregion dimensionhistorical SQL snippet+4 more

Generated SQL

SELECT
  region,
  COUNT(*) AS order_count,
  SUM(amount) AS sales_amount
FROM orders
WHERE created_at >= date_trunc('month', now() - interval '1 month')
GROUP BY region
ORDER BY sales_amount DESC;

Safety Check Passed

Read-onlyAllowed tablesRow-level policyTime range limit

Result Preview

region	order_count	sales_amount
East China	1280	1,294,567.00
South China	950	987,654.00
North China	860	765,432.00

TRUSTED TECHNOLOGY

N Next.js
N NestJS
L LangGraph
P PostgreSQL
R Redis
S SSE

Multi-source Sessions

Unify SQLite, MySQL, PostgreSQL, CSV, and Excel while binding every session to the right data source.

RAG Semantic Context

Hybrid retrieval, RRF fusion, and reranking inject business terms, schemas, historical SQL, and relations into generation.

LangGraph Orchestration

clarify -> retrieve -> planning -> generate-sql -> safety-check -> execute -> format-answer.

Fail-Closed Governance

Read-only SQL, table permissions, workspace isolation, and deny-by-default controls keep data access safe.

Streaming UX & Tool Calls

SSE streams tool-call, tool-result, tool-error, and answer events with observable progress.

Traceable & Replayable

runId binds trace steps, delivery evidence, and replay so every generation can be reviewed.

Domain-aware modeling

Help database tables understand the business domain

Beyond schemas and fields, Text2SQL organizes business terms, entity relations, historical SQL, and data assets into reasoning-ready semantic context.

Business Term Mapping

Map orders, users, merchants, and GMV to real tables and fields.

Relationship Modeling

Add semantic entity links on top of primary and foreign keys.

Query Memory

Feed high-quality SQL and metric definitions back into domain knowledge.

Generation Injection

selected_context carries schemas and business meaning into SQL reasoning.

text2sql

Retail Domain · Model View

Synced 2 minutes ago

Selected Context

{
  "domain": "Retail domain",
  "concepts": ["Orders", "Customers", "GMV"],
  "tables": ["orders", "customers", "payments", "refunds"],
  "relations": true,
  "business_rules": ["GMV = order amount - refund amount", "Only paid orders are counted"]
}

Business Glossary

Orders	Transactions completed by customers through merchants
GMV	Total paid transaction value
Repurchase	Share of customers with at least two orders in a period
Refunds	Returned value from after-sales workflows

History SQL

GMV trend for the last 30 daysFrequent
Order conversion by merchantFrequent
Customer repurchase analysisFrequent

Architecture

An intelligent data analysis system driven by four domains

Every step from user question to executable SQL result is traceable, governed, and explainable.

View full architecture

Frontend

Web AppLive Console

API Gateway

RESTSSE

Conversation Runtime

ClarifyRetrieve KnowledgePlanningGenerate SQLSafety CheckExecute SQLFormat Answer

Knowledge & Governance

WorkspacePermissionsRAG PipelineAudit TrailMetadata Store

Delivery

StreamingQuery ResultTrace / Replay

Workflow

A standard path from question to answer

Clarify

Understand and clarify intent

Retrieve Knowledge

Fetch relevant knowledge and context

Planning

Plan the SQL generation strategy

Generate SQL

Produce candidate SQL

Safety Check

Validate safety and permissions

Execute SQL

Run read-only queries

Format Answer

Return an explainable answer

RAG Intelligence

Three retrieval routes for sharper generation

Lexical, dense, and graph retrieval combine through RRF fusion and reranking, then inject precise selected_context into SQL generation.

Explore RAG details

User Question Orders and revenue by region last month?

Lexical Search BM25

0.82

Dense Search Embedding

0.91

Graph Search Relations & rules

0.88

RRF FusionRerankContext InjectionSQL Generation

Governance

Safety governance that denies by default

Workspace

Datasource binding keeps sessions isolated.

Table-level

Permissions control which tables can be queried.

Policy Version

Rules are versioned for rollout and review.

Audit Trail

Audit events record the full query path.

Fail-Closed

Unknown or risky requests are denied by default.

Roadmap

Continuous evolution for enterprise data assistants

We will keep expanding the capability boundary for safer, more extensible data analysis.

MCP Tool Ecosystem
Connect MCP tools for more internal and external automation.
2026 Q3
Embedded Web Bot
Embed data Q&A into websites, admin panels, and BI pages.
2026 Q4
More Data Connectors
Add warehouses, object storage, and internal APIs.
2027 Q1
ChatBI Artifacts
Generate tables, charts, SQL previews, and diagnostic panels.
2027 Q2
Eval & Rollout Gates
Evaluate SQL accuracy, safety, and rollback risk before release.
2027 Q3

From demo to governable data assistant

Built for developer and team analytics workflows: safe, controllable, and extensible.

Read Docs