AI / LLM Dev Tools
Choose JSONL or JSON array input, then run the check. Expand a record to inspect it; exports include all matching records including those not yet shown.
Preview shows the first 4,000 characters. Copy and download include every selected record.
Input stays in this browser. No upload or network request is made.
Text: 2,000,000 UTF-16 units; UTF-8 files: 20 MiB; arrays: 500,000 units. One record: 500,000 units; 100,000 records maximum.
Structured exports reject selected invalid records and skip explicitly ignored blank lines. Raw export keeps record contents and normalizes separators to LF. Output limit: 40,000,000 units. Array input is split into compact element records first, so raw export in array mode also writes one compact element per line.
Input or options changed. Inspect again before exporting.
A dataset import fails on line 843, but the surrounding records look fine. The JSONL viewer checks each physical line independently so you can locate the broken entry without losing the rest of the file.
This NDJSON viewer handles event logs and line-oriented datasets. A JSON Lines parser checks grammar; readable record previews let you inspect nested values before deciding what to export.
Errors stay beside their original line numbers. Filter to invalid records, read the raw content, and keep those lines in a raw export for investigation. One malformed record does not prevent other records from being checked.
Convert JSONL to JSON or split a JSON array to JSONL while retaining the original numeric tokens and duplicate members. The conversion changes separators and whitespace; it does not rebuild your values as JavaScript objects.
A single array on one line is a valid JSONL record. Select JSON array mode only when you want its top-level elements split into separate records. This choice prevents accidental flattening of an array-valued event.
Use the check button or Ctrl/Cmd+Enter. The JSONL viewer runs manually, including after edits or option changes. An outdated result cannot be copied or downloaded; a cancelled scan cannot replace a newer result.
Records are revealed in batches of 50. Export uses the full filtered collection, not only the visible entries. Choose valid records explicitly before a structured export if the file contains bad lines; raw output can retain those lines.
The JSONL viewer distinguishes invalid lines from valid values with warnings. A repeated name or a number outside JavaScript’s safe integer range deserves attention without being silently rewritten.
Combine a case-insensitive text substring with a status filter. Search looks at source text, so a field name or literal fragment can locate records even if that record cannot be parsed.
Expand individual records into an indented view instead of reformatting the entire source into a document that is no longer JSONL. Physical line numbers remain attached to the original records.
Keep raw line content for debugging, produce compact JSONL for line-based readers, or wrap valid selected records in a JSON array. Each export reports its scope, including explicitly ignored blank lines.
Choose JSON array as the output format after checking the input. All selected non-ignored records must be valid. If any selected line is invalid, filter explicitly to valid records or use raw output to preserve the faulty lines.
Both names describe newline-delimited JSON values. This tool reads .jsonl and .ndjson using the same line rules. Blank-line handling is an explicit option rather than an assumption based on the extension.
Yes. Each line may contain any complete JSON value, including an array or null. In JSONL mode, an array on one line remains one record. Use JSON array input mode to split an entire top-level array.
The record keeps its physical line number, raw text and error information. Later records are still checked. A raw export can include it, but compact JSONL and JSON array output require every selected non-ignored record to pass.
The file limit is 20 MiB. Files are read in chunks, while record and output budgets still apply. The JSONL viewer makes no unlimited-file promise; for larger data, split the source first or use a streaming command-line pipeline.
Unexpected result?
Start with the sample to see how one malformed line remains traceable. Then select the records and output shape needed by your next tool.