Bulk decisions on your data — no code
Upload a CSV, TXT, or JSONL file, pick a judge — one fixed set of Jev questions — and get a structured answer for every row. Watch results land live, pause anytime, and export to CSV or JSONL.
Import your data
Drop a CSV, TXT, or JSONL file. Up to 1,000 rows, 2 MB, and 8,000 characters per row. A CSV with a header row turns each row into an object keyed by column name; a single-column CSV or TXT file treats each line as plain text. Rows are parsed in this tab only — export before leaving.
Drop a file here or click to browse
.csv · .txt · .jsonl — max 2 MB / 1,000 rows| # | Input row |
|---|
Choose a judge
A judge is one fixed set of questions applied to every row — the same Choice, Yes/No, and Score types as the playground and API. Build and test a custom judge in the playground, export the config, and import it here.
Run the batch
Waiting for a file.
Results
Results live in this browser tab. Every row bills actual input tokens — rows rejected before a run and rows that fail are not charged. For recurring or larger jobs, use the Jev API.
How a batch works
1 · Import rows
Drop a CSV, TXT, or JSONL file — up to 1,000 rows and 2 MB. A CSV header row becomes field names, so judges can read each row as a structured object. Files never leave this tab until you run.
2 · Pick a judge
A judge is a fixed set of typed questions — choice, yes/no, or score — run against every row. Start from a preset (triage, moderation, routing, RAG eval and more) or import a playground config you designed and tested on single inputs.
3 · Run and export
Rows process in small chunks with live progress. Pause and resume without losing work, retry failed rows, then export every answer — with probabilities, scores, and token usage — as CSV or JSONL.
Batch, playground, or API?
Playground
One text at a time. Fastest way to design and sanity-check a judge before you point it at a file.
Open the playground →Batch
This page. One-off jobs up to 1,000 rows, zero code, CSV/JSONL in and out. Each row bills only its actual input tokens.
API
Recurring, larger, or real-time workloads. POST the same judge shape to /v1/systemone from your own pipeline — no per-request row limit.
Read the API docs →What file formats does Jev batch accept?
CSV, TXT, and JSONL up to 2 MB and 1,000 rows. A CSV with a header row turns each row into a JSON object keyed by column name; a single-column CSV or TXT treats each line as plain text; each JSONL line is parsed as a JSON value. One row can be at most 8,000 characters once serialized.
What is a judge?
A judge is one fixed set of typed questions — choice, yes/no (noul), or score — applied to every row, the same question types the Jev playground and API use. Pick a preset judge or design your own in the playground, export the config, and import it here.
How is a batch billed?
Each row bills its actual input tokens — the same per-run pricing as the playground, with no batch surcharge. Rows rejected before running and rows that fail are not charged. Guests get 10,000 input tokens; signing in raises the batch throughput limits, and token packs start at $9.90 for 15,000,000 tokens that never expire.
Can I pause or resume a batch?
Yes. Batches run in chunks, so you can pause at any point and resume later — finished rows are kept in this browser tab. Rows that fail are not retried automatically (to avoid double-charging), but you can retry them once you fix the cause.
When should I use the API instead of batch?
Batch is for one-off jobs up to 1,000 rows with no code. For recurring jobs, volumes beyond 1,000 rows, real-time decisions, or integrating results into a pipeline, use the Jev API directly — it has no per-request row limit.
Are my files uploaded to a server?
Your file is parsed locally in this browser tab. When you run a batch, each row is sent to Jev for judging and appears in your run history. Export your results before leaving — the tab is the only place the full data lives.