viewparquet
BlogOpen appOpen
Local by default. Cloud when you want it.

Open Parquet.
Understand what's inside.

ViewParquet turns Parquet into a working surface in seconds: rows, schema, DuckDB SQL, and AI in one browser tab. Start locally with no account. Add only the files you choose to private Cloud projects when you want them on another device.

Your file stays on this device unless you choose Add Cloud copy

Default/flights-200k.parquet
ResultsSource table
delay
distance
time
1
14
405
0.017
2
−11
370
5.5
3
5
389
5.667
4
−5
337
6
5
3
303
6
6
5
236
6.083
7
−4
405
6.167
8
−2
188
6.25
231,083 rows

One dependable path from file to answer

Open instantly
no install and no account required
Stay local
files remain on this device by default
Query properly
full DuckDB SQL in the browser
Continue anywhere
private Cloud copies when you choose

01 · Organize

Projects organize the work. Cloud only changes where a file is available.

Every file belongs to a project, even when it never leaves this browser. Add a private Cloud copy only when you want the same file on another device; the device copy remains independent and fast.

On this device: immediate, private, and free.
Available in Cloud: an explicit private S3 copy.
Open the local appCompare Cloud plans
Default/flights-200k.parquet
ResultsSource table
delay
distance
time
1
14
405
0.017
2
−11
370
5.5
3
5
389
5.667
4
−5
337
6
5
3
303
6
6
5
236
6.083
7
−4
405
6.167
8
−2
188
6.25
231,083 rows

02 · Query

Turn the file into a working answer.

Write full DuckDB SQL with multi-tab queries, schema autocomplete, and one-click execution. Use joins, CTEs, and window functions, then keep the result beside the source that produced it.

SQL
Query 1
Run
SELECT origin,
AVG(delay) AS avg_delay,
COUNT(*) AS flights
FROM data
GROUP BY origin
ORDER BY avg_delay DESC LIMIT 5;
ResultsQuery 1
origin
avg_delay
flights
ORD
12.84
12,768
SFO
11.32
11,904
JFK
10.71
13,491
3 rows

03 · Inspect

See the shape of the data, not just the first page.

Move through large result sets in a virtualized grid, inspect types and column statistics, find values in loaded rows, or filter the complete dataset with DuckDB.

ResultsSource table
delay
distance
time
1
14
405
0.017
2
−11
370
5.5
3
5
389
5.667
4
−5
337
6
5
3
303
6
6
5
236
6.083
7
−4
405
6.167
8
−2
188
6.25
231,083 rows

04 · Ask

AI that works from the same trusted context.

Ask a question in plain language and get auditable SQL, results, and explanations grounded in the open file. Use managed AI with Cloud or connect your own OpenRouter key.

ChatChat 1
Which airports have the highest average delay?

I'll group flights by origin and compare their average delay.

Ran a DuckDB query5 rows
SELECT origin, AVG(delay) AS avg_delay
FROM data GROUP BY origin
ORDER BY avg_delay DESC LIMIT 5;

ORD has the highest average delay in this file, followed by SFO and JFK.

Ask anything about your data…
DeepSeek V4 Flash

05 · Control

Nothing leaves by accident.

Local processing runs in your browser with DuckDB-WASM. Cloud copies are explicit and independent from device copies. Remote files stream from their source, and optional AI sees only the context allowed by your sharing settings.

Add a Cloud copy

Choose where flights-200k.parquet should be available. The copy on this device stays unchanged.

Project

Default
Only this file is copied. Other files on this device remain private until you explicitly add a Cloud copy.
CancelAdd Cloud copy

Remote data

Query the object where it already lives.

Open a public HTTPS link, presigned URL, or s3:// object without copying it into ViewParquet Cloud. DuckDB streams the byte ranges it needs, while private-bucket credentials stay in this browser.

Read the S3 guide
Open from S3 / URL
URLCredentials
Object URL
s3://analytics-prod/events/2024/trips.parquet
Amazon S3Cloudflare R2GCSMinIO

DuckDB streams byte ranges from the bucket; credentials sign requests in-browser and are never sent to viewparquet.

Parquet guides

Start with the task you need to finish

Direct, practical guides for opening, inspecting, converting, and querying Parquet files.

Open & view

Open a Parquet file online

View rows, schema, metadata, and compression without Python or an install.

Convert

Parquet to CSV

Filter the rows and columns you need, then export a spreadsheet-friendly CSV.

Convert

Parquet to JSON

Preserve nested structs and lists in a local JSON Lines export.

Excel

Open Parquet in Excel

Choose the quickest route through CSV export or Microsoft Power Query.

Cloud

Query Parquet on S3

Read public, presigned, or private objects directly from your browser.

Spatial

View GeoParquet

Inspect geometry columns, GeoParquet metadata, attributes, and SQL results.

Need a quick answer? Browse the Parquet FAQ.

Formats

Parquet first. The useful neighbors included.

Purpose-built for columnar files, with GeoParquet, CSV, and JSON available in the same local DuckDB workbench when the job crosses formats.

  • Parquet.parquet

    Native columnar

    First-class
  • GeoParquet.geoparquet

    With geometry

    First-class
  • CSV.csv

    Auto-detected schema

    Supported
  • TSV.tsv

    Tab-delimited

    Supported
  • JSON.json

    Objects or arrays

    Supported
  • JSON Lines.jsonl

    One record per line

    Supported

FAQ

Questions, anticipated.

Got questions? Here are the most common ones about viewparquet.

Do my files leave the browser?

Local datasets are processed with DuckDB WASM and are not uploaded unless you explicitly add a Cloud copy. Remote reads go directly from your browser to the bucket or URL. Optional AI sends only the context you allow, either to your BYOK provider or through managed AI when you choose Cloud mode.

Can I run SQL on my data?

Yes. The SQL editor supports DuckDB syntax with multi-tab queries, schema autocomplete, and exports. Device queries stay in the browser. Paid plans can also run validated read-only queries against Cloud files in isolated, metered workers.

How does the AI assistant work?

Attach one file, several files, or projects to a workspace-wide conversation. Use your own OpenRouter/provider key for direct BYOK mode, or managed AI with a Cloud plan. Sharing controls determine which schema, metadata, profile, or bounded result values can be sent. Managed requests require Zero Data Retention routing.

What file types are supported?

Parquet and GeoParquet are first-class formats for the fastest path. CSV, TSV, JSON, and JSON Lines are also supported as secondary formats.

Is there a file size limit?

Device files are limited by browser, memory, and storage constraints. Cloud supports up to 512 MB per file on Cloud and 2 GB per file on Cloud Pro. A selective query can work on a much larger Parquet source than a full materialization, but plan and result limits still apply.

Does ViewParquet upload my local files?

No. Opening, viewing, or querying a device file does not upload it. A private S3 copy is created only after you explicitly choose Add Cloud copy. Device and Cloud copies can then be removed independently.

How are Cloud queries charged?

Paid plans include a hard monthly infrastructure-cost budget shown as an approximate default-worker minute allowance. Validation and worker-launch failures are not charged. Once a worker starts, cancelled, failed, and successful jobs consume actual billable runtime. There are no automatic overages.

Can I open files from S3 or a URL?

Yes. Open data from a public HTTPS or presigned URL, a public s3:// object, or a private bucket using in-browser access keys (or temporary STS tokens), plus S3-compatible stores like Cloudflare R2, Google Cloud Storage, and MinIO. DuckDB streams only the bytes it needs; credentials stay in your browser and never pass through viewparquet servers.

Working with Parquet files? See the full Parquet FAQ : how to open, view, query, and convert Parquet.

Changelog

See what's new in viewparquet
2.7

Shared agent conversations and isolated Cloud queries

2.6

Local-first projects, with Cloud when you want it

2.5

Copy cells and keep your column layout

2.4

BYOK AI assistant for private data work

Learn & Explore

Latest Blog Posts

AI data pipelines, Parquet for LLM workloads, DuckDB, and columnar analytics.

View all articles
AILLMFeatured

How to Use AI with Parquet Without Uploading Your Data or Buying Another AI Plan

Three practical ways to analyze Parquet with an LLM: let a browser-capable agent operate ViewParquet, bring your own mod…

7/28/20269 min read
Read more
S3ParquetFeatured

How to Open Parquet from S3 Privately in Your Browser

Query Parquet on AWS S3, Cloudflare R2, GCS, or MinIO without uploading files to a third-party viewer. viewparquet strea…

6/24/20269 min read
Read more
ParquetDatasetsFeatured

Free Public Parquet Datasets You Can Download Right Now (Every Link Verified)

A curated list of 11 public Parquet files you can actually download — flights, NYC taxi trips, Hugging Face NLP sets, an…

6/12/202610 min read
Read more