About viewparquet

Trust & methodology

About viewparquet

Last reviewed: August 19, 2026

viewparquet is a browser-based workbench for inspecting and querying Parquet data without installing a desktop data stack. This page records who is responsible for the product, where data travels, what the current viewer does not do, and how we qualify technical claims.

Who makes viewparquet

viewparquet is built by Antony Adhiban and operated by Kaladev Digital Private Limited. Articles name their author and show their publication, update, or verification date when one is available. Product changes are recorded in the changelog.

Three distinct data paths

“In-browser” applies to the core viewer, not to every optional connection a user can make. The distinction matters:

Local files
Parquet, GeoParquet, CSV, TSV, JSON, and JSON Lines files opened from your device are processed by DuckDB-WASM in the browser. Their contents are not uploaded to viewparquet servers for viewing or SQL execution.
Remote files
For HTTPS, presigned, or S3-compatible sources, the browser connects directly to the host. viewparquet does not proxy or cache the dataset. Compatible Parquet endpoints can serve range requests, but actual transfer depends on the query, file layout, codecs, and host. Private object-store credentials remain in the current browser session.
Optional AI
AI is BYOK-only. When you invoke it, your browser sends your message and applicable context directly to the OpenRouter or supported direct provider you configured. Settings control preloaded schema, profile fields, and samples; chat tools and saved AI context may also provide structural metadata or SQL. viewparquet does not proxy AI requests.

The Privacy Policy documents analytics, error reporting, local storage, support, and the AI data path in more detail.

What the viewer supports today

  • Open local Parquet, GeoParquet, CSV, TSV, JSON, and JSON Lines files.
  • Open supported formats from public HTTPS or presigned links, public S3 objects, or private AWS S3 and compatible endpoints including R2, GCS with HMAC interoperability, and MinIO.
  • Browse rows, inspect the loaded schema and metadata, and run browser-supported DuckDB SQL.
  • Export a source view or query result as Parquet, CSV, or JSON Lines in the browser.
  • Optionally use a BYOK assistant to draft, explain, or repair SQL.

Current boundaries

  • There is no guaranteed maximum file size or query time. Browser and WebAssembly memory, the device, file layout, codecs, endpoint behavior, and query shape all affect the result.
  • GeoParquet is exposed as a table with stored geometry values. The viewer does not currently render a map, decode geometry into GIS features, or validate GeoParquet CRS and metadata.
  • The viewer does not currently create .xlsx files. CSV and JSON-family exports can reinterpret or lose type and display fidelity, so verify important conversions.
  • Remote access cannot bypass CORS, object permissions, an incorrect region, or expired authorization. Range reads also depend on the host and the Parquet/query layout.
  • AI output can be incomplete or wrong. Review generated SQL and results before relying on them for production, financial, legal, medical, or other consequential decisions.

How claims are checked

  1. Start with shipped behavior. Capability and privacy copy is compared with the current implementation and exercised in the current viewer. Automated checks guard especially easy-to-overstate claims about uploads, remote range reads, file size, exports, GeoParquet, and optional AI.
  2. State conditions and limits. Performance is not reduced to one file-size promise. Remote behavior names its CORS, authorization, endpoint, file-layout, and query dependencies.
  3. Date external verification. Links and third-party datasets can change. The public Parquet dataset guide records when its files were downloaded and the exact DuckDB checks used. A dated result is evidence for that run, not a permanent availability guarantee.
  4. Prefer reproducible evidence. Benchmarks and compatibility claims should name the input, environment, procedure, raw result, and verification date so someone else can repeat them.

Corrections and feedback

If a claim is stale, a sample no longer opens, or the product behaves differently from this page, please use the feedback board or the contact options on the Support page. Include the page URL, browser, file type, and a reproducible example when it is safe to share. Do not send credentials or sensitive datasets.