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
Multi-step reasoning + hybrid retrieval (keyword/vector/semantic), 14 layers of SQL-injection defense.
Auto-infers the best chart type and produces interactive charts — no BI tool setup.
Conversational weekly/monthly reports, scheduled delivery to DingTalk, WeCom, Slack, Lark.
Threshold and anomaly detection with cross-channel alerts (13+ messaging channels).
RAG document QA + knowledge graph, federated across PostgreSQL, MySQL, Oracle, ClickHouse, MongoDB.
Side by side
| Dimension | Traditional BI tool | Hand-written SQL | DataWeaver agents |
|---|---|---|---|
| Time to answer a new question | A model and a report must exist first; a new metric means reconfiguration | File a ticket with IT and wait 2–3 days | Ask directly; result and chart come back in seconds |
| Querying across databases | Usually needs ETL into a warehouse before anything can be joined | Connect to each database, export, then align definitions by hand | Federated queries across PostgreSQL, MySQL, Oracle, ClickHouse and MongoDB |
| Skill required | Requires drag-and-drop modelling and metric definition skills | Requires SQL plus knowledge of the schema | Plain language is enough; the Viz agent infers the chart type |
| Security and ongoing monitoring | Permissions live in the BI platform; anomalies still rely on someone watching dashboards | Injection defense and alert scripts are re-implemented per project, hard to standardise | NL2SQL ships 14 layers of injection defense; the Monitor agent alerts across 13+ messaging channels |
Quantified before / after
Verifiable facts
The product is built from five specialized agents — NL2SQL, Viz, Report, Monitor and Knowledge — covering querying, visualization, reporting, monitoring and knowledge QA.
The NL2SQL agent implements 14 layers of SQL-injection defense and uses hybrid retrieval across keyword, vector and semantic-summary indexes.
The Monitor agent dispatches threshold and anomaly alerts to 13+ messaging channels; the Report agent supports scheduled delivery to DingTalk, WeCom, Slack and Lark.
Federated querying spans five database families: PostgreSQL, MySQL, Oracle, ClickHouse and MongoDB.
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.
Where it fits
Further reading