I built Spektr because I needed analytics across different datasets (finance, Jira exports, project tracking) without sending data to the cloud. It's a Go-based analytics engine with a CLI. Point it at any CSV, ask questions in English, get charts and tables back. The AI (Gemini) only sees your column names and schema — never your actual data. Try it: spektr --file your-data.csv --query "show totals by category" What it does under the hood: auto-detects your schema (dimensions vs measures), translates your question to a structured query via Gemini, runs all computation locally, returns charts/tables/text summaries. Zero-copy RecordView interface means it works with any data shape without materializing intermediate objects. Works as a Go library, CLI binary, or WASM/npm module.

Built in Go. MIT licensed. Happy to answer questions about the architecture.