Python integration

The CLI is the primary interface. This API is available for embedding jevotron in an existing application or ETL process.

from pathlib import Path
from jevotron import Config, scan
from jevotron.parsers import CSV

chunks = CSV(id_column="ident")(Path("examples/airports/spiked.csv"))
for result in scan(chunks, Config(guidance="Check geographic consistency.")):
    if result.warning:
        print(result.id, result.score)

scan(chunks, config, cache=..., refresh=False, client=None, batch_size=1, batch_tokens=48000, stats=None) is a generator. batch_size="auto" chooses an economical layout; an integer caps entries per request, and 1 selects the original per-entry layout. preview accepts the same batching options and shows planned requests assuming cache misses. See batching.

Pass a jevotron.runner.ScanStats() instance as stats to collect new API submissions, actual input/output tokens, and estimated input tokens. These counters are per request, so batched entries do not multiply token totals. Pass cache=None to disable persistence. An injected client implements evaluate(request) -> response using the Jev HTTP JSON contract. This makes offline testing and external integrations straightforward. If stopping a scan early, close its iterator (or use contextlib.closing) to release resources. The caller owns an injected client's lifetime.