Privacy Policy

Local files stay on your device

Parquet, GeoParquet, CSV, and JSON files you open are processed locally in your browser using DuckDB-WASM. We do not upload a local dataset for viewing or SQL execution unless you explicitly choose Save to cloud or upload it to a cloud project. You can verify this in your browser's developer tools Network tab: file reads come from disk, a remote URL, or object storage, not from viewparquet servers.

Optional ViewParquet Cloud

Cloud is an optional, account-based service. When you explicitly save a file to cloud, your browser uploads a copy directly to a private Amazon S3 bucket using a short-lived signed URL. We store the file name, size, content type, project association, and object key in our database. Opening a cloud file creates a short-lived signed download URL so DuckDB-WASM can read it in your browser. Files already stored locally are never synchronized automatically.

Supabase processes account identity and project metadata. Amazon Web Services stores cloud file objects. You can delete individual cloud files from the project library; account deletion and enterprise retention requests can be made through Support.

Remote sources (S3 and URLs)

When you open a dataset from HTTPS or S3 (including private buckets), your browser connects directly to that host. DuckDB-WASM streams byte ranges from the object; viewparquet does not proxy or cache your files on our servers.

Access keys and session tokens you enter for private buckets are used only inside your browser to sign requests. They are not persisted on our servers. Query-free public URLs may be saved locally as recents. URLs containing query parameters, including presigned URLs, are kept only for the current browser tab; reopening a private source prompts you to enter credentials again.

Optional AI assistant

AI features are optional. Local users can bring their own OpenRouter or supported direct-provider key. Cloud subscribers can use managed AI without creating a provider account.

In managed Cloud AI mode, your authenticated request is sent to a viewparquet server endpoint and then to OpenRouter. The request can include your messages and the dataset context allowed by your sharing settings. We record the model ID, input/output token counts, estimated cost, account ID, and timestamp to enforce the plan allowance; we do not intentionally store prompt or response bodies in the usage ledger.

In BYOK mode, chat and SQL-helper requests are sent directly from your browser to the provider you configure, typically OpenRouter. viewparquet does not proxy BYOK requests through our servers, does not receive your API key, and does not log prompts, schema, generated SQL, responses, or result previews by default.

Depending on your Settings → AI data-sharing choices, requests may include:

• Your messages and SQL prompts
• Column names and types from the open file
• Dataset metadata (file name, row count, column count, file size)
• Bounded query-result values when that toggle is enabled
• Value-bearing profile facts or SQL-helper samples only when you enable those settings

Chat and SQL helper can use different sharing settings. Use the presets (Minimal, Balanced, Full context) or fine-tune each toggle. Do not include secrets or personal data you are not permitted to share with your chosen model provider.

You may optionally add direct provider keys (for example Anthropic) in advanced settings. Keys are kept in session storage for the current tab by default. You can explicitly opt in to remembering them in local storage on the device. They are sent only to the provider you select.

No cookies, no cookie banner

We use privacy-friendly analytics to understand aggregate product usage, such as page views and feature interactions. We do not use analytics cookies, advertising trackers, or cross-site profiling for this purpose. Because nothing is stored on your device for tracking, there is no cookie banner to click through.

Information we may collect

• Cloud account and project data: email, account identifiers, subscription status, project names, and metadata for files you explicitly store in cloud.
• Billing: Polar processes checkout, tax calculation, invoices, and subscription management. We receive customer, product, subscription, and payment-status identifiers but not full card details.
• Product analytics (cookieless): we use Vercel Analytics to collect aggregated page views, referral sources, and feature usage outside the viewer. It is cookieless and does not receive your file contents or AI prompts.
• Error monitoring (essential): we use Sentry to capture crash and error reports so we can keep the app working. This is part of the core service and does not include your file contents, AI prompts, or automatic personal identifiers such as your IP address.
• Support messages: when you contact us through the in-app messenger, Featurebase processes your message and conversation metadata so we can reply. Do not include sensitive datasets or secrets in support messages.
• Local storage: preferences (such as theme and AI settings), chat session history, and optionally saved file contents in IndexedDB on your device only. BYOK keys use session storage by default and use local storage only when you enable “Remember keys on this device.” These are functional settings, not trackers. BYOK API keys never leave your browser except when sent to the AI provider you configure.

Third-party services

The app may load open-source runtime assets (for example DuckDB-WASM) from our CDN. Cloud services use Supabase for authentication and metadata, AWS S3 for private object storage, Polar for billing, and OpenRouter for managed model access. BYOK mode communicates with the provider you choose. See OpenRouter's Privacy Policy. Vercel Analytics processes data according to Vercel's Privacy & Compliance policy, Sentry according to Sentry's Privacy Policy, and Featurebase according to Featurebase's Privacy Policy.

Contact

Questions about this policy? Visit our Support page or reach us through the channels listed there.