SlothDB is a from-scratch C++20 embedded SQL database in active development. Query Parquet, CSV, JSON, Avro, Arrow, Excel, and SQLite files directly with SQL, in-process. Early-stage; read the Status section on GitHub before treating any performance numbers as final.

pip install slothdb
python -c "import slothdb; slothdb.demo()"

Generates a 100 000-row CSV, runs three queries, and prints the side-by-side with DuckDB shown above. No files to find, no setup.

import slothdb
db = slothdb.connect()
df = db.sql("SELECT region, SUM(revenue) FROM 'sales.parquet' GROUP BY region").fetchdf()

No server. No import step. No CREATE TABLE. Point SQL at files on disk.

• Four benchmark-fitted shortcuts deleted from the engine. Two were returning wrong results on inputs outside the benchmark. The previous "33 of 43 beats DuckDB, up to 5×" claim was removed alongside them. Public discussion of the engine's architecture: ClickHouse/ClickBench#930. Treat any remaining performance numbers as anecdotal.
• DATE and TIMESTAMP columns render as ISO strings (2013-07-02, 2013-07-15 12:40:00) instead of raw epoch integers.
• slothdb_version() and the shell --version report the release tag.
• 424 doctest cases pass.

• Nested aggregates work everywhere. ROUND(AVG(x)), AVG(x) + 1, SUM(x) / COUNT(*), CAST(SUM(y) AS DOUBLE) and similar shapes that wrap an aggregate inside a scalar function or arithmetic used to throw "Function execution for: AVG". Fixed.
• ORDER BY by aggregate alias works. SELECT region, COUNT(*) AS cnt ... ORDER BY cnt DESC no longer silently sorts by column 0.
• Arithmetic type promotion fixed. AVG(x) + 1 no longer drops the +1 and AVG(x) / COUNT(*) no longer returns inf.
• 408 unit tests, 131,537 assertions, green on Windows, Linux, macOS.

Same embedded model as DuckDB and SQLite. You link it into your process and point SQL at files. Different defaults:

• 7 file formats built in - Parquet, CSV, JSON, Avro, Arrow, SQLite, Excel. DuckDB needs extensions for Avro and SQLite.
• Stable C ABI. Numeric error codes don't shift between releases. Bindings built against 0.1.x keep working.
• ~1-4 MB single binary, fully self-contained.

Performance vs DuckDB is competitive on some queries and worse on others. The GitHub README's Status section has the architecture context behind the numbers.

import slothdb

# In-memory
db = slothdb.connect()

# Query files directly
db.sql("SELECT * FROM 'data.csv' WHERE score > 90").show()
db.sql("SELECT COUNT(*) FROM 'logs.parquet'").show()
db.sql("SELECT * FROM read_json('events.json') LIMIT 5").show()
db.sql("SELECT * FROM sqlite_scan('app.db', 'users')").show()

# Persistent database
db = slothdb.connect("analytics.slothdb")

# DataFrame integration
df = db.sql("SELECT region, SUM(revenue) FROM 'sales.csv' GROUP BY region").fetchdf()

• No multi-writer transactions (single writer, crash-safe checkpoint).
• No distributed execution. Single-node embedded engine.
• No secondary indexes. Scan-based execution; zone-map pruning helps on sorted data, but no B-tree / hash index for point lookups.
• Window-function coverage is partial. Plain OVER / PARTITION BY works; ROWS BETWEEN ... frames and cumulative SUM OVER (ORDER BY) shapes have known gaps.
• Authenticated S3 not implemented. s3:// URLs work for anonymous public-bucket reads only.
• Some SQL corners still surprise you. Open an issue with a repro.
• 0.2.x, about a year old. Treat as beta.

Performance varies query-to-query: SlothDB is competitive or faster on some shapes (native Avro decode, CSV COUNT(*), small joins) and slower on others (high-cardinality multi-column GROUP BY, the generic non-handler path). The GitHub README's Status section explains the architectural reason and what's being changed.

pip install slothdb && python -c "import slothdb; slothdb.demo()" runs a 3-query side-by-side against whatever DuckDB version is installed on your machine, so you can see the comparison locally.

There's a Discord: discord.gg/XJWyGmX5G. Bug reports, install help, weird query plans, "is this slower than it should be", feature ideas - any of it. The maintainer reads everything. GitHub issues are still the canonical tracker; the server is for the questions that come before you file one.

• Source: https://github.com/SouravRoy-ETL/slothdb
• Discord: https://discord.gg/XJWyGmX5G
• Changelog: https://github.com/SouravRoy-ETL/slothdb/blob/main/CHANGELOG.md
• Issues: https://github.com/SouravRoy-ETL/slothdb/issues
• SQL reference: https://github.com/SouravRoy-ETL/slothdb/blob/main/docs/DOCUMENTATION.md

MIT