DuckDB.

DuckDB is an in-process analytic database — like SQLite, but tuned for analytical queries on columnar data. Orchid opens any .duckdb file, and you can query Parquet, CSV, and JSON files directly without loading them first.

Requirements

• A .duckdb file on your machine, or just the files (Parquet/CSV/JSON) you want to query — DuckDB doesn't need a database file at all to read external files.
• Read permission on the file(s).

Connect

1. Open Integrations → + Add connection → DuckDB.
2. Either:
   • Click Browse and pick a .duckdb file, or
   • Leave the path empty for an in-memory database (queries on Parquet/CSV files work directly).
3. Click Save. The schema appears immediately — no test step required.

Querying files directly

DuckDB's killer feature for an analyst's notebook: read external files with no load step. Use the file path right in the FROM clause.

-- Query a single Parquet file
SELECT user_id, count(*) AS events
FROM '/Users/me/data/events-2026-05.parquet'
GROUP BY user_id
ORDER BY events DESC
LIMIT 100;
      

-- Glob across an entire directory of Parquet files
SELECT date_trunc('day', event_ts) AS day,
       count(*) AS events
FROM '/Users/me/data/events-*.parquet'
WHERE event_ts > now() - interval 30 day
GROUP BY 1
ORDER BY 1;
      

-- CSV with options
SELECT *
FROM read_csv_auto('/Users/me/data/customers.csv',
                   header = true,
                   sample_size = -1);
      

-- JSON
SELECT name, address.city AS city
FROM read_json_auto('/Users/me/data/customers.ndjson');
      

Optional settings

S3 / cloud storage

DuckDB can query Parquet directly from S3, GCS, or any S3-compatible store. Configure credentials inside a cell once per session.

-- One-time per session: install + load the httpfs extension
INSTALL httpfs;
LOAD httpfs;

-- Configure S3 credentials
SET s3_region = 'us-east-1';
SET s3_access_key_id = '...';
SET s3_secret_access_key = '...';

-- Now query S3 directly
SELECT count(*) FROM 's3://my-bucket/events/2026/05/*.parquet';
      

Persistent vs in-memory

Pointing the connection at a .duckdb file gives you a persistent database — any tables you create stay. Leaving the path empty starts an in-memory session that resets every time you restart Orchid. In-memory is perfect for ad-hoc work on external Parquet files where the database itself is just a query engine.

Common gotchas

• "file does not exist" — paths in SQL strings are case-sensitive on macOS and Linux. Check the literal path on disk.
• "HTTP 403 / SignatureDoesNotMatch" when reading from S3 — credentials wrong or region mismatch. Set s3_region to match the bucket's region.
• Mismatched Parquet schemas in a glob — if your Parquet files have evolving schemas, DuckDB throws on schema mismatch. Use union_by_name = true in the parquet_scan options to align by column name.
• Slow queries on a single large CSV — convert to Parquet once for big speedups: COPY (SELECT * FROM read_csv_auto('big.csv')) TO 'big.parquet' (FORMAT PARQUET);
• Database file locked — DuckDB allows a single writer process. If your local app has the file open in write mode, Orchid's connection will fail. Close the other process.

Example queries

-- List tables in the current database
SELECT table_name FROM information_schema.tables WHERE table_schema = 'main';
      

-- Pivot Parquet data into a daily summary
SELECT
  date_trunc('day', event_ts) AS day,
  count(*) FILTER (WHERE event = 'signup') AS signups,
  count(*) FILTER (WHERE event = 'login')  AS logins,
  count(DISTINCT user_id)                  AS daily_active_users
FROM '/Users/me/data/events-2026-*.parquet'
GROUP BY 1
ORDER BY 1;
      

Where to go next

For more on writing SQL cells, see SQL cells.