JEV / BATCH

Jev batch decisions

Apply the same questions to a file of inputs. Import CSV, TXT or JSONL, choose a preset or your own questions, and export structured results. Processing runs in your browser while the page stays open.

Open batch workspace Read the docs Jev Batch Classification: CSV & JSONL

CSV, TXT or JSONL · 2 MB · 1,000 rows. Serialized state and the complete request must each fit 8,000 characters, including question text. Multi-column CSV uses the first row as headers.

Test your questions first. Each successful row uses one free run or actual input-token credits.

Pause between requests, retry failed rows, and export CSV. Keep the page open until processing finishes.

Input file examples

Upload CSV, TXT, or JSONL. A multi-column CSV uses its first row as headers; each later row becomes one state object. A single-column CSV is treated as a list of strings and keeps its first row. TXT uses one non-empty line per state; JSONL uses one JSON object, array, or string per line.

issue,region
"Charged twice; please refund order A1042",EU
"The app crashes on launch",US

Shared questions and row state

Choose a preset or define questions and criteria once; the same questions are sent for every row. Each parsed row replaces request state: CSV/JSON objects stay objects, while TXT and single-column CSV rows stay strings.

Output and export

CSV exports use exactly these columns. The answer and error cells contain the corresponding row result; JSONL export writes one result object per line. These examples describe the file format, not model performance.

row,input,status,answer,error

File and request limits

The file limit is 2,000,000 bytes (2 MB) and 1,000 rows. Each serialized state and the complete request, including all question text and criteria, must separately stay within 8,000 characters.

Run, pause, retry, and keep results

Pause takes effect after the current row finishes. You can retry failed rows and export results as CSV or JSONL. Processing runs in the current browser tab; closing or reloading the page stops it, and unexported batch results are not promised to resume automatically.

Each successful row follows the existing usage rules: one free run when available, then the applicable input-token credits. Loading a file or example does not run the model.