What's new

Workspace-wide AI conversations

• Attach files or projects with @ mentions and keep the conversation independent of the open file
• Use the same multi-file context with Local BYOK AI or managed Cloud AI
• Edit an earlier prompt, keep branches, and sync eligible conversation history across devices
• Choose draft-only, confirm-before-run, or automatic read-only query permission

Cloud query execution

• Run Cloud-only files directly from private S3 without downloading them into the browser first
• Use scale-to-zero Fargate workers with hard runtime, concurrency, result, and monthly-cost limits
• Receive realtime job updates, bounded previews, and seven-day Parquet result downloads
• Keep device DuckDB as the default whenever every source is locally available

Launch privacy and security

Managed AI requires Zero Data Retention routing. Each query worker receives only one short-lived job token and exact-object signed URLs, while conversation and query-content retention is enforced automatically.

Projects for every workflow

• A Default project keeps first-time and existing local workflows immediately usable
• Move files between projects without changing where their data is stored
• See Device and Cloud availability on one logical file instead of duplicate entries

Optional Cloud copies

Local viewing remains free, private, and account-free. Adding a Cloud copy is always an explicit action, and removing it does not remove the copy on your device.

• Resume interrupted large uploads and avoid storing duplicate file contents
• Open a Cloud-only file on a new device and keep a fast device copy for later use
• Keep workspace layout preferences consistent after signing in on another device

Managed AI and billing

Cloud plans can use managed AI without setting up OpenRouter. Local BYOK remains available, and higher-cost models simply consume AI credits faster.

Cell selection and copy

• Select a single cell or drag across a rectangular range
• Copy from the context menu or with the operating system keyboard shortcut
• Choose spreadsheet-compatible, plain-text, or header-inclusive output

Persistent column layouts

• Drag column headers to reorder the grid without changing source data
• Keep column order and widths independently for each source table and query result
• Restore the original column layout from the subtle customized-layout control

Discoverability

A new Help & shortcuts menu in the viewer header documents table gestures and operating-system-specific keyboard shortcuts.

Runtime reliability

DuckDB and the SQL editor now load their executable runtime assets directly from viewparquet, keeping the viewer available under the production Content Security Policy without relying on third-party script CDNs.

BYOK-only AI

• OpenRouter key setup in Settings -> AI -> Connection
• Chat and SQL helper requests go directly from the browser to the configured provider
• No server-side AI proxy flow

Model control

Users can pick from curated OpenRouter models, hide defaults, and add custom model slugs for their own provider account.

Privacy controls

Data-sharing presets control what schema, metadata, and sample values can be sent to the selected BYOK provider.

Remote sources

• Open from S3 / URL in the viewer — public HTTPS, presigned URLs, or s3:// paths
• Private buckets with in-browser access keys and optional STS session tokens
• Provider presets for Amazon S3, Cloudflare R2, Google Cloud Storage, and MinIO
• Recent sources saved without secrets for quick reopen of public links

Privacy

Remote reads go directly from your browser to the bucket or CDN. viewparquet does not proxy object data or store your credentials on our servers.

AI chat assistant

• Ask questions in plain language and get SQL grounded in your open file
• Run generated queries or insert SQL into the editor in one click
• Multi-tab chat sessions while you explore
• Charts and query-tool results inline in the conversation

SQL editor AI

• Ask AI (⌘K / Ctrl+K) to draft queries from a prompt
• Fix AI to repair syntax errors against your schema

Bring your own key

Connect an OpenRouter API key in Settings → AI. Keys stay in your browser — viewparquet does not proxy AI requests through our servers. You choose the model and pay your provider directly.

Data sharing controls

• Presets and per-feature toggles for column headers, types, metadata, and sample cell values
• Separate settings for chat vs SQL helper
• Context chip in chat shows what is shared at a glance

Support

• In-app messenger via Featurebase for help and feedback

Privacy-first analytics

We replaced invasive analytics with a cookieless product-analytics boundary. It records aggregate page and allowlisted feature activity without filenames, SQL, prompts, values, or viewer identifiers.

• No analytics cookies, so no cookie banner
• No ad trackers, user profiles, or session recordings
• Local files remain in the browser unless the user explicitly adds a Cloud copy

SQL editor

Write full DuckDB SQL with tabbed queries, schema-aware editing, and one-click run. Export results to Parquet, CSV, or JSON without leaving the tab.

Results grid

• Virtualized Glide grid for smooth scrolling on large result sets
• Type-aware cell coloring for numbers, dates, IDs, and nulls
• Per-column stats: cardinality, ranges, and null rates

Privacy-first

Local files do not upload for ordinary viewing or SQL. Optional Cloud copies were introduced later as an explicit action.

Viewer

• Remember recently opened files in IndexedDB for quick reopen
• Improved Parquet and GeoParquet detection and error messages
• Dark mode aligned with the rest of the app

Formats & export

• GeoParquet viewing alongside standard Parquet
• Export query results to CSV and JSON
• Clearer schema and column type display in the grid

Core

• Client-side Parquet viewing powered by DuckDB-WASM
• No accounts, no uploads, no server-side data processing
• Support for .parquet and .geoparquet files