Question 1

Does SelectStar send my database data to a third party?

Accepted Answer

SelectStar runs inside your own Next.js deployment. Only schema metadata, query text, and small data previews are sent to the LLM provider you configure — and you can run that provider on-prem with Ollama, vLLM, or LM Studio. Large result sets are cached server-side by id; only the id and a preview travel through the LLM context.

Question 2

What's the difference between SQL Mode and Classic Mode?

Accepted Answer

SQL Mode connects to a live database (SQLite, PostgreSQL, MySQL-ready) via a connection string. Classic Mode lets you upload CSV, TSV, XLSX, XLSM, XLSB, or ODS files (up to 25 MB each). Each file is parsed into in-memory SQLite tables — one table per XLSX sheet — and the agents treat them like any database. You can JOIN across tables from different files and edit cells directly in the spreadsheet grid.

Question 3

How does the graph orchestrator differ from a ReAct loop?

Accepted Answer

A ReAct loop gives the LLM a list of tools and lets it repeatedly output Thought → Tool Call → Observation until it's done. That's flexible but unreliable for databases. SelectStar uses a graph with conditional edges: the Router classifies intent, then deterministic edges route to the right agents. EDA, Viz, and ML run in parallel after a SELECT. Synthesis is the only agent that writes prose.

Question 4

What is Zen mode and why does it matter?

Accepted Answer

Zen mode is the write-safety layer. It's off by default. When Zen mode is on and the SQL agent produces a write statement, it is NEVER auto-executed. It's registered as pending (10-minute TTL) and surfaced to the user with an impact estimate. The user sees three actions: Confirm & execute, Dry-run (rollback), or Cancel. Every resolution is audit-logged.

Question 5

How can the ML agent work without Python?

Accepted Answer

All the math is implemented in pure TypeScript. OLS regression uses normal equations + Gaussian elimination. k-means uses k-means++ initialization. Linear-trend forecasting computes slope and R². No scikit-learn, no Python runtime, no extra dependencies.

Question 6

Is SelectStar free?

Accepted Answer

Yes — SelectStar is MIT-licensed and free forever. You bring your own LLM, so the only cost is whatever inference you already pay for. There is no hosted SaaS tier and no usage-based pricing.

Built in the open

Four ways to engage

GitHub

Issues

Discussions

Contributing

The basics

Built with a focused, modern stack

Try SelectStar, then star it on GitHub