Private data tables in your browser
For analysts and developers who need to eyeball a CSV or Parquet without uploading it anywhere, Local Analyst is a private data table that runs entirely in your browser. Unlike hosted AI tools, there are no accounts, no uploads, and no server that sees your rows — even our demo cannot see them.
How it works
- You drop a file: CSV, Parquet, JSON, XLSX or Avro.
- Your browser opens it with DuckDB compiled to WebAssembly — a full analytical SQL engine running on your device.
- Profiling, charts, pivot, JOIN and SQL all run locally. Exports (CSV/Parquet/HTML report) are generated locally and downloaded by you.
- Our server only delivers the static app files (HTML/JS/WASM). It never receives your data, because the app never sends it.
Proof you can verify: load the sample data, then disconnect Wi-Fi. Everything keeps working.
Privacy: what we store and what we don't
- No accounts, no tracking cookies. There is nothing to log into and nothing that follows you.
- In your browser (localStorage/OPFS only): theme choice, AI consent flag, saved queries. All of it stays on your device.
- On our server: anonymous download counts from web-server logs (page loads per day, referrer). IP addresses are truncated and kept at most 24 hours solely to estimate unique visitors. No fingerprints, no cross-site identifiers, nothing sold.
- Local AI (optional): the app ships zero model weights. If you enable it, you download the model yourself (file or URL) and it runs in your tab. Prompts sent to the on-device model contain schema and aggregates only — never raw rows.
- Delete everything: clear site data for this domain in your browser settings. Since nothing of yours is on our server, there is nothing to request.
Limits (honest)
- Times are shown in UTC — hover any date cell for the exact ISO value. Click a TIMESTAMP column badge to see raw delivered values instead of dates.
- Browser memory ceiling is roughly 2–4 GB depending on device. Huge files are streamed from disk automatically — or convert to Parquet first, it scans faster.
- The engine is single-threaded by default; multithreading behind
?threads=1disables some file-format extensions — a known DuckDB-WASM tradeoff. - Local AI models are small (0.5–1.5B parameters). They draft SQL that you must review — the Run button always stays in your hands.