RowCoveAccount ↗

Arrow and Feather.
View the useful records.

Arrow IPC and Feather · Local desktop workspace

Open supported Arrow IPC and Feather files in RowCove on Windows and macOS. Inspect field types, select columns and browse local records with a native Rust engine. Supported bounded sources also offer staged edits and Save copy.

Download for Windows ↓Download for macOS ↓

Free Preview: up to 100 read-only records. No account or card.
Version 0.1.1 · Windows x64 · macOS Apple Silicon · Install guide →

How do I open an Arrow or Feather file without uploading it?

Choose Open data in RowCove and select the file with the native picker. Review the detected reader and schema, then inspect a read-only first page of up to 100 records in free Preview. Pro adds the supported query and output workflows shown for that source.

Open Arrow IPC or Feather on your desktop

Training features, intermediate analysis tables and market-data exports often need a quick visual check before another job starts. Open the file directly instead of writing a one-off inspection script or uploading a sensitive dataset to a web service.

RowCove detects the source from file evidence and shows its available operations. A file extension alone cannot establish that a damaged file or an unsupported variant is readable. If detection fails, review the reported reader and error before trying another workflow.

Keep field types and complete values in view

Review names, types and representative records before assuming a column's meaning. Check timestamp units, missing values and identifiers; a number displayed in a table is not automatically a price, date or category. Select a cell to inspect its complete value rather than relying on clipped text.

The following examples are synthetic mock field choices, not customer data or a required training schema.

Arrow inspection choices · synthetic (mock) data
TaskSelected fieldsMock scope
Review a feature rowrecord_id, feature_score, splitmock_001 · demo split
Inspect a market recordsymbol, timestamp, priceDEMO-A · one interval
Check missing valuesrecord_id, labelsynthetic examples

Choose columns before expanding the task

Project the fields needed for the question and use supported Pro filtering, sorting and result pages. Keep the selected source and result scope with your findings. A quick first page does not imply that a full scan or a complete sort has the same cost.

Arrow and Parquet have different access and storage characteristics. The Parquet viewer and editing workflow explains its column and partition choices; the format matrix lists reader-specific operations. Published 20 GB mock CSV timings measure CSV, not Arrow throughput.

Stage supported changes and save a separate copy

For a source that offers editing, activate Pro, switch to Edit and review the cell changes. Save copy writes a new artifact while preserving the original. Reopen the output and check the changed values, field types and representative records before passing it to another tool.

Parquet and Arrow edit-copy currently require a source no larger than 128 MiB, with at most 200,000 rows and 512 columns. Larger files can still support inspection and queries without being eligible for editing. Dataset-folder browsing is read-only: keep the source-file provenance, then open an eligible individual shard to review a correction. Check the columnar editing workflow before buying Pro for this task.

Inspect Arrow dataset folders with a clear scope

Choose a supported dataset folder when records are split across Arrow shards. Review the discovered files and recognized partitions, choose the intended split and fields, then continue through bounded pages while a cursor is present. Preserve source-file provenance with returned records.

An empty filtered page can still carry a continuation cursor. A folder sample does not establish a global row count, complete schema reconciliation or a global sort. Follow the Parquet and Arrow shard guide and pretraining dataset workflow for source selection and continuation.

Keep local processing and AI sharing deliberate

Dataset processing runs locally on your device. Account activation and updates use network services. When you connect an AI assistant through RowCove MCP, requested results go to that client and may reach its model provider.

Authorize only the intended folder and request selected fields and bounded results. Use the MCP setup guide for permissions and Preview and Pro plans for access requirements. CSV and JSONL editing remain available through their own format-specific workflows.

Try the first page on your own data.

Start on Windows or macOS: source information and up to 100 records, read-only. Review reader capabilities before planning a full-file Pro workflow.

Download for Windows ↓Download for macOS ↓

Version 0.1.1 · Windows x64 · macOS Apple Silicon · Install guide →

Reader capabilities →

Arrow and Feather workflows

Questions, answered

Contact support →
Will editing overwrite my original dataset?

Sources open read-only by default. Where the reader supports editing, switch to Edit and stage your changes. Save copy writes the result to a separate destination, keeping the original available for comparison and recovery.

Before saving a large result, check the destination and free disk space. Open the new copy and verify representative records, types and the intended changes. Some readers are read-only or expose metadata rather than editable records; use the operations shown for the detected source.

Follow the save-copy workflow Check editing support

How do I open JSONL, Parquet or Arrow data split across many files?

Use the folder picker or drop a dataset folder into Open data. For supported collections, RowCove discovers shards and available split or partition information. Choose the relevant source scope, project the fields you need and browse bounded result pages with source context.

Keep the scope explicit when comparing results: one split, a date partition and the entire collection are different populations. Follow the continuation state through pages when you need every matching record. A bounded collection scan should not be treated as an automatic global sort, aggregate or complete schema reconciliation across every shard.

Browse a pretraining shard collection Work with columnar datasets

Do my datasets ever have to leave my computer?

You do not need to upload a dataset to RowCove to inspect, query or edit it. Those desktop operations run on your device. Account activation and billing use separate online services for information such as your purchase email, subscription and device access; opening a dataset does not send its contents to those services.

MCP is an explicit sharing path: requested tool results go to the AI client you connect. If that client uses a cloud model, results may reach its provider. Choose the folders, fields and result size you intend to share, and review the client’s own privacy policy before using sensitive data.

Read the privacy policy Control MCP sharing