Show HN: DataTalk CLI, Query CSV and Excel in Plain English Using LLM and DuckDB (github.com)

DataTalk is a new CLI that lets you query CSV, Excel (.xlsx/.xls) and Parquet files in plain English using an LLM-driven interface with DuckDB as the local execution engine. Rather than writing SQL or juggling awk/csvkit flags, you run dtalk <file> and ask natural-language questions (interactive or single-prompt). It’s pitched as fast, scriptable, and privacy-first: DuckDB performs local analytics (multi‑gigabyte files are supported, Parquet preferred for speed), only the table schema is sent when using cloud LLMs, and you can run completely offline by pairing with local Ollama models. Technically, DataTalk uses LiteLLM to unify 100+ model providers (OpenAI, Anthropic, Google, Ollama, etc.), exposes generated SQL (--sql/--sql-only) for transparency, and outputs human-readable tables or machine-friendly JSON/CSV for pipelines. Installation is pip install datatalk-cli (Python 3.9+); configuration is via LLM_MODEL and provider API keys (or Ollama for offline). Key implications: it lowers the barrier to ad-hoc data exploration, fits into automation workflows with exit codes and structured outputs, and offers a pragmatic privacy tradeoff by keeping raw data local while leveraging LLMs for query synthesis.