seeparquet

seeparquet is an interactive terminal browser for Parquet files. It follows the navigation and rendering conventions of seej while using PyArrow to read Parquet row groups in bounded batches.

Features

• Browse records with the arrow keys or Enter
• Start at an absolute row index without scanning skipped row groups
• Filter records with safe expressions using the ex variable
• Select a subset of columns
• Inspect the Arrow schema and Parquet metadata
• Reuse seej renderers and renderer plugins
• Automatically unpack and render gzip/tar data stored in task_binary
• Print a fixed number of rows in non-interactive scripts

Install

pip install seeparquet

From a source checkout, for development:

pip install -e '.[dev]'

Usage

# Interactive browsing
seeparquet data.parquet

# Start at row 1000
seeparquet data.parquet -I 1000

# Read only selected columns
seeparquet data.parquet -C id,messages

# Filter rows; repeat -c to combine expressions with AND
seeparquet data.parquet -c 'ex["status"] == "active"'
seeparquet data.parquet -c 'ex.score >= 0.8' -c 'len(ex.messages) > 2'

# Use a seej renderer for selected fields
seeparquet data.parquet -r messages=messages,conversation

# Inspect metadata and schema
seeparquet data.parquet --schema

# Print three matching rows and exit
seeparquet data.parquet --head 3 -R

# Decode task_binary into metadata.json, instruction.md, and other files
seeparquet tasks.parquet -C path,task_binary

# List seej renderers, including installed plugins
seeparquet --list-renderers

Keyboard shortcuts

• Down, Right, or Enter: next matching row
• Up or Left: previous loaded row
• T: toggle rich/raw rendering
• S: show Parquet schema and metadata
• Q or X: quit

Filtering happens after rows are decoded. Column projection and row-index skipping are pushed into the Parquet reader.