What is SelectStar?
SelectStar is an open-source agentic database-analysis platform. You connect a database or upload a spreadsheet, ask a natural-language question, and a team of specialized AI agents — coordinated by a graph orchestrator — writes the SQL, profiles the data, builds the charts, and runs the models. You just read the answers.
Think of it as sitting between a BI tool, a Jupyter notebook, and a chat assistant — but agentic, so it decides the right type of analysis on its own instead of you having to pick a chart type or write SQL manually.
The four things to understand
Graph orchestrator, not a ReAct loop
Six specialized agents (Router, Schema, SQL, EDA, Viz, ML, Synthesis) coordinated by a graph with conditional edges. The Router classifies intent up front, then deterministic routes decide which agents run. EDA, Viz, and ML fork-join in parallel after a SELECT. No unbounded reasoning loops.
Zen mode — write safety
Read-only by default. When Zen mode is on and the SQL agent produces a write, it is NEVER auto-executed. It's registered as pending (10-minute TTL) and surfaced with an impact estimate. The user sees three actions: Confirm & execute, Dry-run (rollback), or Cancel. Every resolution is audit-logged.
Two modes — SQL and Classic
SQL Mode connects to a live database (SQLite, PostgreSQL, MySQL-ready) via a connection string. Classic Mode parses uploaded CSV, TSV, XLSX, XLSM, XLSB, or ODS files into in-memory SQLite tables — one table per XLSX sheet. The agents treat them identically. You can JOIN across files and edit cells directly in Classic Mode.
Pure-TypeScript ML — no Python
The ML agent runs OLS regression (via normal equations + Gaussian elimination), k-means clustering (with k-means++ initialization), and linear-trend forecasting (with R² and slope metrics). All in pure TypeScript. No scikit-learn, no Python runtime, no extra dependencies. Scoped to 'quick triage' — escalate to a real notebook for deep work.
Ready to try it?
The fastest path is the installation guide — you'll be asking questions in under a minute.
Go to installation