Jev AIJev workspace
Upload a file, choose a judge and get a Jev decision for every row: probabilities, labels and scores you can export.
These questions are applied to every row in your file.
Jev batch processing applies one set of typed questions to every row of a file. Label support tickets, moderate comments or score reviews without writing code.
Drop in a CSV, TXT or JSONL file of up to 2 MB and 1,000 rows, with up to 8,000 characters per row. Check the preview before you run anything.
A judge is a set of yes/no, choice and score questions. Start from a demo, pick one from My judges, or describe your task and let AI draft one from your first rows.
Rows run one by one and results appear as they finish. Pause whenever you like, then download CSV or JSONL with every answer, its probabilities and token usage.
Any decision you can ask about one piece of text, you can ask about a whole file.
CSV, TXT and JSONL files encoded as UTF-8, up to 2 MB and 1,000 rows. In a TXT file each line is one input. In a JSONL file each line is a JSON object or array. A CSV file with a header row turns each row into an object keyed by column name. Without a header, a single-column row becomes plain text.
Each row costs 1 credit, the same as one playground run. If your credits run out, the row is charged by its actual input tokens instead. Generating a judge with AI also costs 1 credit. Rows rejected before Jev runs are not charged.
Rows run one at a time with a short pause between them, so 100 rows take roughly five to six minutes. You can pause at any point and continue later from the same row.
Your file is parsed in the browser and kept only in the current tab. Each row is sent to Jev when it runs and appears in your run history like any other run. Export your results before you close the tab.
Yes. Choose a demo judge, a judge saved under My judges, or describe what you need and let AI draft one from your first three rows. The judge is locked when the batch starts, so every row is answered by the same questions.
The batch stops at that row and shows the error. Rows rejected before execution, for example because of a low balance or rate limit, go back to pending and can be retried. Rows with an uncertain outcome are marked failed and never repeated automatically, so you are not charged twice.
Use batch processing for one-off jobs of up to 1,000 rows without writing code. For recurring jobs, larger volumes or real-time decisions, call the Jev API from your own backend with the same questions.