ChatBI that treats SQL generation as a multi-agent team sport

What it does

Aix-DB is a conversational BI (ChatBI) web application that sits between business users and SQL databases. You type a question in natural language—Chinese or English—and it parses the intent, retrieves relevant schema context via RAG, generates SQL, runs it against MySQL, PostgreSQL, Oracle, ClickHouse, or several Chinese warehouses like DM and StarRocks, and returns an ECharts or AntV visualization. The whole flow is orchestrated through LangChain and LangGraph with an MCP-based multi-agent “Skill” layer handling different stages.

The interesting bit

The project treats SQL generation not as a single prompt but as a collaborative pipeline where different agents handle intent parsing, schema retrieval, SQL drafting, and validation. It also explicitly targets the Chinese enterprise stack—supporting Dameng (DM), Apache Doris, and StarRocks alongside the usual Western suspects—which is rarer than it should be in open-source BI tools.

Key highlights

• Supports 8+ database backends including MySQL, PostgreSQL, Oracle, ClickHouse, DM, Apache Doris, StarRocks, plus CSV/Excel upload
• Uses a hybrid RAG retriever (Embedding + BM25) backed by Neo4j graph knowledge to ground schema context
• Exposes both a Vue 3 web UI and a CLI tool (published to npm) for different user workflows
• Packaged as a Docker image with Sanic serving the API gateway and JWT authentication built in
• Offers a “Skill mode” that appears to let users register domain-specific agent capabilities, though the README is light on implementation details

Caveats

• The README is heavy on paid support tiers, WeChat QR codes, and custom-development pitches, which suggests the open-source repo may serve partly as a lead magnet
• Several marketing claims (“越问越准” / “more accurate the more you ask”) are stated without benchmarks or evaluation data to back them up
• The CLI section and MCP Skill architecture details are truncated or vague; expect to read source code to understand how the agent collaboration actually works

Verdict

Worth a look if you need an on-premise ChatBI prototype for a Chinese-centric data stack and don’t mind filling in documentation gaps. Skip it if you require battle-tested accuracy guarantees or a community-driven project without commercial strings attached.