What it does

Upload an Excel file or connect a database, type “最近三个月销售额趋势如何？” in a browser tab, and this agent streams back SQL generation, query execution, chart selection, and a business summary via SSE. It’s a local Flask app (port 5001) aimed at analysts who’d rather not write SQL or wrestle with BI tools.

The interesting bit

The chart catalog is unusually deep for an open-source project—43 types including Marimekko, ridgeline plots, chord diagrams, and horizon charts. The agent auto-recommends one based on query results, which suggests someone actually thought about visualization semantics rather than slapping a bar chart on everything.

Key highlights

• Multi-source: Excel/CSV, SQLite, MySQL, PostgreSQL, SQL Server; v4.0 switched internal processing to DuckDB for “万行数据秒处理”
• 43 chart types across 6 categories (comparison, time, distribution, geospatial, relationship, part-to-whole)
• Slash commands for direct SQL (/sql), forced charting (/chart), stats (/analyze), K-means (/kmeans), decision trees (/tree), and export to Word/PPT (/report, /ppt)
• LLM-agnostic: DeepSeek, OpenAI, Claude, or any OpenAI-compatible API via configurable base URL/model/key
• MCP support for extending agent skills with local or remote Model Context Protocol servers
• Knowledge base uploads to ground analysis in business-specific context
• Time series module added in v4.0: Prophet, SARIMA, ARIMA, VAR, GRU

Caveats

• The “one-click install” scripts are explicitly marked as unstable and in testing
• Generated chart links break on restart because they’re stored in local outputs/charts/—the README acknowledges this in the FAQ
• Requires bringing your own API key; no local model inference

Verdict

Worth a spin if you need to hand non-technical colleagues a self-hosted tool that answers business questions with actual statistical depth. Skip it if you need production-grade access control, collaborative editing, or already have a mature BI stack your users tolerate.