Skip to main content

AWK + stdlib + Rust

The AWK you know,
for the data you have now.

zawk runs your AWK programs, compiled with Cranelift, and adds CSV, JSON, Parquet, Excel, databases, S3 and NATS as input, ~190 standard library functions, modern syntax, and tooling built for AI agents.

Get StartedWhy zawk?
brew install linux-china/tap/zawk
$ zawk '$FI["age"] > 20 { print $FI["name"], $FI["age"] }' \
    'mysql://root:***@127.0.0.1:3306/test#people'
Jackie 48
Love 48
Leo 24

$ zawk -i parquet 'NR <= 2' people.parquet
{"id":1,"name":"Jackie","age":48}
{"id":2,"name":"Love","age":48}

$ echo hello | zawk '{ print ulid(), encode("base64", $0), mask("abc@example.com") }'
01M60KH8PG6NM4HE8ZKQKXSAXT aGVsbG8= abc*********com

Reads and writes

  • CSV / TSV
  • JSON Lines
  • Parquet
  • Excel
  • Markdown
  • SQLite
  • MySQL
  • PostgreSQL
  • S3
  • HTTP
  • NATS
  • gz · zst · xz

Everything AWK was missing

Fast by design

Type inference and a Cranelift JIT turn AWK into machine code, with SIMD CSV parsing and parallel execution. Usually 2-8x faster than gawk on large files.

zawk -i csv -p r -j 8 '{ s += $3 } END { print s }' big.csv

Structured data, by name

RFC 4180 CSV, TSV, JSON Lines, Parquet, Excel and Markdown tables, with columns by name through $FI["col"]. Write CSV, JSON Lines, Parquet or Markdown.

zawk -i jsonl -o md '{ print $FI["name"], $FI["age"] }' users.jsonl

A real standard library

~190 functions: text, JSON/XML/HTML, dates, crypto and JWT, IDs, statistics, HTTP, email and S3. Failures warn and set ERRNO instead of crashing.

{ print url($1)["host"], json_value($2, "$.id"), uuid("v7") }

Databases as input

Read a SQLite, MySQL or PostgreSQL table or query like a CSV file, and write back with bound parameters. Redis and NATS KV are one call away.

zawk '{ print $FI["name"] }' 'postgres://u:p@host/db#people'

Streams and messages

Subscribe to a NATS subject as input, then publish to NATS or MQTT. A ten-line AWK script becomes a message processor.

zawk -i json '{ publish(out, MSG_BODY) }' 'nats://localhost:4222/logs.>'

Made for AI agents

zawk llm writes the prompt, zawk lint checks the answer as JSON, and jev asks an LLM typed questions per record. A Claude Code skill is included.

zawk llm -i csv -H sales.csv 'top 3 regions?' | claude -p

Modern syntax, still AWK

f-strings, raw strings, ??, try … else, match expressions,local variables, key-value loops, ranges, !in and negative field indexes. Existing AWK programs keep their meaning.

Explore the syntax →
{ print f"{NR:>4}: {toupper($2):<10} {$3 * 1.1:,.2f}" }
{ sql = f"SELECT * FROM {table} WHERE id = {$1}" }

Start with a one-liner

Install zawk and point it at a file you already have.

Read the docsBrowse tutorials