csvql-query 1.5.5

Fast CSV querying from Python — powered by a Zig/SIMD engine

csvql-query

Query CSV files with SQL from Python — powered by a Zig/SIMD engine.

Zero-copy mmap reads + SIMD parsing happen before Python ever sees the data. Faster than DuckDB on typical workloads, no dependencies required.

Installation

pip install csvql-query

Quick Start

import csvql

# Returns a list of dicts (like csv.DictReader, but with SQL)
rows = csvql.query("SELECT name, salary FROM 'employees.csv' WHERE salary > 100000 ORDER BY salary DESC")
# [{'name': 'Alice', 'salary': '185000'}, ...]

# Raw CSV string
csv_str = csvql.query_csv("SELECT * FROM 'data.csv' LIMIT 10")

# pandas DataFrame (pandas must be installed)
df = csvql.query_df("SELECT category, COUNT(*) as n FROM 'sales.csv' GROUP BY category")

# (headers, rows) tuples — no dependencies
headers, rows = csvql.query_tuples("SELECT name, age FROM 'users.csv' WHERE age > 25")

API

Function	Returns	Description
query(sql)	list[dict]	Execute SQL, get list of dicts
query_csv(sql)	str	Execute SQL, get raw CSV string
query_df(sql)	DataFrame	Execute SQL, get pandas DataFrame
query_tuples(sql)	(list[str], list[tuple])	Execute SQL, get (headers, rows)

SQL Support

The SQL path is embedded in the query string (same as the CLI):

# Filtering, ordering, limiting
csvql.query("SELECT name, city FROM 'data.csv' WHERE age > 30 ORDER BY name LIMIT 5")

# Aggregation
csvql.query("SELECT department, AVG(salary) FROM 'emp.csv' GROUP BY department")

# Unix pipes — use '-' as the filename
import subprocess, sys
# or just pass stdin data via the engine directly

Full SQL reference: SIMPLE_QUERY_LANGUAGE.md

Performance

• mmap + SIMD parsing — data is never copied into Python memory
• Parallel chunk processing on multi-core machines
• Typically 5–9x faster than DuckDB on 1M-row CSVs

Requirements

• Python ≥ 3.10
• macOS (x86_64 / arm64) or Linux (x86_64)
• pandas optional — only needed for query_df()

Links

• GitHub
• CLI Installation
• SQL Reference