Duckdb Analytics Engine

DuckDB & Polars High-Performance Analytics Skill (duckdb-analytics-engine)

Enables AI agents to execute sub-second columnar SQL queries over large local datasets (well logs, financial ledgers, web access logs, Parquet files) directly in memory with zero database server overhead.

Execution Patterns

1. Querying Local Parquet / CSV / JSONL via DuckDB

import duckdb

# Query multiple Parquet/CSV files directly using SQL
query = """
SELECT 
    formation_name,
    AVG(gamma_ray) AS avg_gr,
    MAX(resistivity) AS max_res,
    COUNT(*) AS sample_count
FROM read_parquet('data/well_logs/*.parquet')
WHERE depth_m BETWEEN 2000 AND 3500
GROUP BY formation_name
ORDER BY avg_gr ASC;
"""

df = duckdb.query(query).to_df()
print(df)

2. High-Speed LAS / Well Log Batch Summaries

import duckdb

# Execute direct SQL aggregations over exported CSV/JSON log data
res = duckdb.sql("""
    SELECT 
        COUNT(*) as total_samples,
        PERCENTILE_CONT(0.5) WITHIN GROUP (ORDER BY gr) as median_gr,
        SUM(CASE WHEN gr < 45 AND res > 20 THEN 0.15 ELSE 0 END) as estimated_pay_m
    FROM 'well_log_export.csv'
""")
res.show()

Best Practices for Federation Agents

1. Sub-second Performance: Use duckdb-analytics-engine instead of raw Python for loops when parsing tabular files > 5MB.
2. Zero Setup Overhead: DuckDB runs in-process without requiring external database services or network ports.