jsonl-tail

Tail JSONL files (eval logs, datasets) with pretty-printed JSON, filters, and live follow mode.

Install

uv tool install jsonl-tail   # or: pip install jsonl-tail

Usage

# last 10 records, pretty-printed
jsonl-tail eval.log.jsonl

# all records, one line each (pipe-friendly)
jsonl-tail -a -c eval.log.jsonl

# watch a live eval log for WARN/ERROR records
jsonl-tail -f --regex '"(WARN|ERROR)"' eval.log.jsonl

# project one key per record
jsonl-tail -k score eval.log.jsonl

Flag	Meaning
-n, --last N	Last N records (default 10)
-a, --all	Print all records
-f, --follow	Wait for appended records (tail -f)
-c, --compact	One-line JSON per record
-k, --key KEY	Print only this top-level key per record
--filter SUBSTR	Only records containing SUBSTR
--regex PAT	Only records matching PAT

Invalid JSON lines pass through raw with a marker — corrupt logs never crash the tool. Set NO_COLOR=1 for plain output.

Development

uv sync
uv run pytest
uv run ruff check .