IndustryEnterprise data / Middleware

DataWeaver

Five specialized agents cut cross-database natural-language queries from 2–3 days to 10 seconds.

In short

DataWeaver is YGG's cross-database analytics middleware for enterprises. A business user asks a question in plain language and five specialized agents — NL2SQL, Viz, Report, Monitor and Knowledge — take turns fetching data, drawing the chart, writing the report and raising alerts, returning an answer in seconds instead of queueing for an IT-written SQL query. What separates it from a traditional BI tool is that it federates queries directly across PostgreSQL, MySQL, Oracle, ClickHouse and MongoDB, with no warehouse or semantic model to build first.

What the agent takes off people's plates

Core capabilities

01

NL2SQL Agent

Multi-step reasoning + hybrid retrieval (keyword/vector/semantic), 14 layers of SQL-injection defense.

02

Viz Agent

Auto-infers the best chart type and produces interactive charts — no BI tool setup.

03

Report Agent

Conversational weekly/monthly reports, scheduled delivery to DingTalk, WeCom, Slack, Lark.

04

Monitor Agent

Threshold and anomaly detection with cross-channel alerts (13+ messaging channels).

05

Knowledge Agent

RAG document QA + knowledge graph, federated across PostgreSQL, MySQL, Oracle, ClickHouse, MongoDB.

Side by side

Cross-database natural-language querying vs traditional BI tools vs hand-written SQL

DimensionTraditional BI toolHand-written SQLDataWeaver agents
Time to answer a new questionA model and a report must exist first; a new metric means reconfigurationFile a ticket with IT and wait 2–3 daysAsk directly; result and chart come back in seconds
Querying across databasesUsually needs ETL into a warehouse before anything can be joinedConnect to each database, export, then align definitions by handFederated queries across PostgreSQL, MySQL, Oracle, ClickHouse and MongoDB
Skill requiredRequires drag-and-drop modelling and metric definition skillsRequires SQL plus knowledge of the schemaPlain language is enough; the Viz agent infers the chart type
Security and ongoing monitoringPermissions live in the BI platform; anomalies still rely on someone watching dashboardsInjection defense and alert scripts are re-implemented per project, hard to standardiseNL2SQL ships 14 layers of injection defense; the Monitor agent alerts across 13+ messaging channels

Quantified before / after

Measured impact

Operations Director
BeforeWaited 2–3 days for IT to build each report
AfterAsks a question, gets a chart in 10 s
Finance Manager
BeforeReconciled three systems manually in Excel
AfterA single federated query returns the answer
CEO
BeforeAsked an assistant to prep 20 charts the day before
AfterAsks the agent on the way in, gets answers instantly
Sales Manager
BeforeNeeded IT to write SQL for conversion analysis
AfterSelf-serve answers, same-day playbook iteration
IT Director
BeforeAlways writing ad-hoc reports for business teams
AfterSelf-hosted open-source freed them for architecture & security

Verifiable facts

Numbers and sources

The product is built from five specialized agents — NL2SQL, Viz, Report, Monitor and Knowledge — covering querying, visualization, reporting, monitoring and knowledge QA.

Self-reportedDataWeaver product documentation, verified 2026-082026-08-01

The NL2SQL agent implements 14 layers of SQL-injection defense and uses hybrid retrieval across keyword, vector and semantic-summary indexes.

Self-reportedDataWeaver product documentation, verified 2026-082026-08-01

The Monitor agent dispatches threshold and anomaly alerts to 13+ messaging channels; the Report agent supports scheduled delivery to DingTalk, WeCom, Slack and Lark.

Self-reportedDataWeaver product documentation, verified 2026-082026-08-01

Federated querying spans five database families: PostgreSQL, MySQL, Oracle, ClickHouse and MongoDB.

Self-reportedDataWeaver product documentation, verified 2026-082026-08-01

The claim "from 2–3 days down to seconds" compares a business team filing an ad-hoc report request into the IT backlog against asking the same question once inside DataWeaver.

Self-reportedMeasured in YGG's own DataWeaver demo environment; not independently verified2026-08-01

Where it fits

When this approach does not apply

  • Accuracy depends heavily on how readable the underlying schema is. On legacy systems with cryptic column names, no comments and no key relationships, NL2SQL hit rates drop noticeably — a semantic layer (table and column glossaries, metric definitions) usually has to be added before go-live.
  • It is a tool for exploratory analysis and day-to-day business questions, not the final authority on externally reported figures. Numbers destined for financial statements, regulatory filings or audits still need a human to review the generated SQL and its definitions before use.
  • Federated query performance is capped by the source databases. Large cross-database joins still push load back onto production; point it at read replicas and cap time windows and row counts, or "fast answers" turn into a slow transactional system.
Written by: YGG Technology solutions teamPublished: 2026-05-08Last reviewed: 2026-08-01