SQLGlot — SQL Parser, Transpiler & Optimizer in Pure Python

What it is

SQLGlot is a pure Python library with no external dependencies that parses SQL into an AST, transpiles queries between 20+ SQL dialects, and optimizes query plans. It supports Snowflake, BigQuery, DuckDB, Spark, Postgres, MySQL, and many more dialects without requiring a database connection.

Data engineers migrating between warehouses, building SQL linters, or creating multi-dialect tools will find SQLGlot essential. It handles the parsing and dialect translation that would otherwise require writing custom regex or maintaining separate query files per database.

How it saves time or tokens

Manually rewriting SQL queries from one dialect to another is error-prone and time-consuming. SQLGlot automates this conversion with a single function call. It also optimizes queries by pushing down predicates and simplifying expressions, which can reduce token counts when SQL is part of LLM prompts.

How to use

1. Install SQLGlot with pip (no extra dependencies needed).
2. Parse SQL strings into AST objects for inspection or transformation.
3. Transpile between dialects by specifying the source and target.

Example

import sqlglot

# Transpile from Spark SQL to BigQuery
spark_sql = "SELECT EPOCH(ts) FROM events"
bigquery_sql = sqlglot.transpile(
    spark_sql,
    read='spark',
    write='bigquery'
)[0]
print(bigquery_sql)
# Output: SELECT UNIX_SECONDS(ts) FROM events

# Parse and inspect the AST
ast = sqlglot.parse_one('SELECT a, b FROM t WHERE a > 1')
for col in ast.find_all(sqlglot.exp.Column):
    print(col.name)

Related on TokRepo

• AI tools for database — Other tools for database management, querying, and migration.
• AI tools for coding — Developer tools that assist with code transformation and analysis.

Common pitfalls

• Assuming 100% dialect coverage. SQLGlot handles the most common syntax differences, but edge cases in vendor-specific extensions may need manual review.
• Forgetting to specify the source dialect. Without read='dialect', SQLGlot uses its default parser which may misinterpret dialect-specific syntax.
• Using SQLGlot as a query executor. It only parses and transforms SQL text; it does not connect to or run queries against a database.