RowCoveAccount ↗Open JSON without pasting it into an online tool
Dataset exports, experiment metadata and configuration records can contain information you want to keep on your own device. Choose Open data, select a .json file in the native file picker, review its sample and open the document. You do not need Python, a notebook or a cloud import to inspect supported records.
Free Preview needs no account or card. It shows a read-only first page of up to 100 records, with complete cell inspection. Pro adds the supported filtering, sorting and edit-copy workflow. The same Pro plan covers supported CSV, JSONL, Parquet and Arrow operations; there is no separate per-format purchase.
RowCove presents JSON as records, rather than a code editor with syntax highlighting. An array supplies multiple records; a single object or scalar supplies one record. This workflow is useful when your question concerns fields and examples, rather than writing arbitrary JSON source code.
Inspect nested fields without losing the record
Nested object fields appear as table columns such as meta.split. Select a cell to read its complete value in the record inspector. Arrays and other structured values need their full context: a short visible excerpt is not the complete training conversation, label set or metadata object.
The following small example is synthetic mock data only.
| record_id | meta.split | label | Review |
|---|---|---|---|
| mock_001 | demo | unreviewed | Inspect the original record |
| mock_002 | demo | reviewed | Compare a saved copy |
Scroll horizontally to inspect all fields. Keyboard: focus the table and use the arrow keys.
Check identifiers, numeric precision, nulls and missing fields before relying on a comparison. Choose a numeric sort only when the field really represents numbers; identifiers and category codes may need text ordering.
Answer a document-wide question, then inspect the matching records
Find loaded rows searches the page already in view. To ask a question about the opened document, use the supported Pro filter and sort controls. Choose the relevant field, set the predicate and inspect the returned scope. Keep the source row with your findings.
For example, filter a mock category or split, inspect missing labels, then order a numeric score before reviewing candidate records. Keep the filter, ordering and document version with your notes so the result can be reproduced. A matching first page alone does not establish anything about a different file or a collection of shards.
JSON documents are parsed within a deliberate size budget. This differs from a line-oriented JSONL stream or projected columnar access. The published 20 GB mock CSV comparison measures CSV; its timings do not establish JSON throughput.
Change a field and save a separate JSON copy
- Open the document and check the intended source record and field.
- Switch to Edit and select the cell.
- Review the complete value, enter the replacement and choose Stage edit.
- Review the staged change; undo it if needed.
- Choose Save copy and a new destination in the native save dialog.
- Reopen the output and verify the changed field, other records and expected types.
Supported nested-field edits are saved back into JSON structure, rather than exporting the flattened table as the document. The source stays unchanged. Saved JSON may use different indentation or formatting, so use it as a data copy rather than a byte-for-byte source-code preservation workflow.
JSON types matter. A valid JSON literal can represent a number, boolean, null, array, object or quoted string. To keep a string that looks like another JSON type—for example the text “true”—enter it as a quoted JSON string and verify the output. Use the getting-started guide for cell inspection, staged changes and safe-copy workflows.
Choose JSON, JSONL or a columnar workflow for the actual file
| Source | Useful workflow | What to know |
|---|---|---|
| JSON document | Inspect a bounded array or object; edit supported fields | 256 MiB document limit; up to two million records |
| JSONL / NDJSON | Browse line-oriented records and supported training-data shards | One JSON record per line; not an arbitrary JSON document |
| Parquet | Select columns and supported partitions | Reader-specific query and bounded editing limits |
| Arrow / Feather | Inspect typed columnar records locally | Supported IPC variant and source capabilities matter |
A document larger than the JSON budget is not made streamable by renaming it .jsonl. Produce a valid record-oriented export using the source system, or choose a suitable columnar export. For pretraining, SFT and preference data, follow the nested training-record guide and post-training data workflow.
Local processing, with deliberate sharing through MCP
Viewing, querying and saving supported documents run on your device. Account activation, billing and updates use network services separately. When an AI client calls RowCove MCP, requested results are shared with that client and may reach its model provider.
Authorize the intended folder and ask for selected fields and bounded records. Review the tool schema and result scope before expanding the task. Start with the MCP setup tutorial and complete format capability matrix.
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.
Version 0.1.1 · Windows x64 · macOS Apple Silicon · Install guide →
Reader capabilities →