Apparently, our newsletter has a longer release cycle than our software. July and August slipped past, and suddenly there’s a summer’s worth of GizmoData news to catch up on.
Since June, we’ve expanded the ADBC driver family, connected GizmoSQL to Claude, made bulk loading dramatically faster, and continued building out our cloud platform. There’s also a new advisor, a maritime application with 18.7 billion records, and a fresh benchmark that finishes before your coffee cools.
Let’s get into it. 👇
🆕 New here? Meet GizmoSQL.
GizmoSQL puts DuckDB behind an Apache Arrow Flight SQL server, so applications, dashboards, notebooks, and agents can share its analytical engine over the network. The core is open source; Enterprise adds production controls and observability. Run it yourself or explore our cloud options. Explore GizmoSQL.
☁️ GizmoData Cloud—and Cloud at Customer
In June, we mentioned our managed service in private preview. There’s now much more to show.
GizmoData Cloud brings GizmoSQL clusters and DuckLake catalogs across AWS, Google Cloud, and Azure into one console. Provision compute near your data, manage catalogs and secrets, configure autoscaling, and stop idle clusters. Cloud and region awareness helps keep compute placement—and data movement—intentional. Hosted Cloud remains in early access. Take the tour.
Cloud at Customer is already deployed in Summation, Company Wrench, and Vaden Automotive’s environments. These installations bring the GizmoData platform into customers’ own infrastructure—with the control plane, portal, and GizmoSQL deployments inside their cloud accounts. It’s a deployment model we’re delivering and supporting with real customers today. One installation can manage multiple clouds. Explore Cloud at Customer.
Underneath it is a useful separation: GizmoSQL supplies compute, while DuckLake maintains the shared transactional catalog and data. Adding another node doesn’t require copying the lakehouse. Chris Wones explains the architecture in “GizmoSQL + DuckLake: Elastic Compute Over Shared State.”
🔌 One ADBC driver became a whole family
July brought a major client release: GizmoSQL ADBC 2.0, rebuilt around a Go native library.
The familiar Python package remains adbc-driver-gizmosql, with a compatible API. The shared core brings GizmoSQL-specific capabilities—including OAuth/SSO, connection profiles, bulk ingestion, and automatic DDL/DML routing—to other ADBC language bindings too.
Python users can upgrade with:
pip install --upgrade adbc-driver-gizmosqlThe ecosystem moved with it: dbt, SQLMesh, QGIS, SQLFrame, and our Node.js client adopted the new driver. Read the ADBC 2.0 story.
Then we expanded the family:
Oracle: a Go driver that connects without Oracle Instant Client.
IBM Db2: a Go driver that connects without IBM CLI or JCC.
DuckDB Quack: our driver for DuckDB’s native remote protocol, alongside GizmoSQL’s Flight SQL driver.
Because these drivers exchange Arrow batches, a source cursor can stream into a destination’s bulk-ingest API. Move data from Oracle or Db2 into GizmoSQL—or back again—without an intermediate dataframe or export file. DuckDB’s ADBC extensions also make SQL-based integration possible. Explore the driver suite and examples.
And a milestone since that announcement: Oracle ADBC reached its first stable release, v1.0.0, on September 1. Release notes.
🚀 Bulk ingestion: nearly 12 million rows in five seconds
GizmoSQL v1.37.0 replaced its per-cell Arrow conversion code with DuckDB’s own vectorized Arrow scanner.
In the published TPC-H lineitem test, loading approximately 12 million rows dropped from 36 seconds to 5 seconds—7.2× faster. That measurement used 50,000-row batches over localhost; the gain depends on the workload, particularly column count and conversion costs.
The rewrite also improves correctness: decimals follow DuckDB’s native conversion path, missing columns receive their defaults, and GeoArrow geometry columns land as actual GEOMETRY columns.
Chris Harrison wrote a walkthrough with a runnable benchmark so you can try both versions yourself. Read the bulk-ingest article.
🤖 Talk to your GizmoSQL data from Claude
We shipped gizmosql-mcp, an open-source MCP server and Claude Desktop extension.
Connect it to GizmoSQL and your assistant can explore catalogs and tables, inspect query plans, and execute parameterized SQL. Configure multiple connections to work across environments in one conversation.
It starts read-only, with query limits and timeouts enforced through GizmoSQL. Claude Desktop stores credentials in the OS keychain. For database-enforced access restrictions, use a read-only token or Enterprise catalog permissions.
It works with Claude Desktop, Claude Code, and other MCP clients—and it’s Apache-2.0 licensed. Get started with GizmoSQL MCP.
We’ve kept refining it since launch, including automatic reconnection after a GizmoSQL restart and adapting session refresh to the server’s idle-timeout configuration. Latest MCP releases.
⏱️ Benchmark corner: TPC-H 1 TB in 80.5 seconds
We tested GizmoSQL on AWS’s new Graviton5 hardware and reran the comparison on Graviton4 and Azure Cobalt 100 using the same DuckDB version.
AWS Graviton5: 80.5 seconds
AWS Graviton4: 89.8 seconds
Azure Cobalt 100: 106.8 seconds
The Graviton5 result represents about $0.17 of query compute at the tested on-demand rate.
These totals sum the mean of three executions of each of the 22 queries. Data staging and ingestion are excluded, and the Azure comparison includes hardware differences beyond the CPU.
The clearest comparison is the AWS generation change: roughly 10% less query time on Graviton5 with the same software. Methodology, costs, and every query timing.
🛠️ Server and ecosystem highlights
The June Dispatch covered GizmoSQL through v1.29. We’re now at v1.38.5, with plenty of improvements beyond the headline features. Server releases.
A few worth calling out:
More places to run: native Windows ARM64 builds and more portable Linux binaries, including support for Raspberry Pi OS and Amazon Linux 2023. v1.30.0
Better operational controls: runtime-adjustable graceful shutdown, concurrent-session limits, and idle-session eviction. v1.32.0, v1.36.0
Clearer query timing: Enterprise instrumentation separates execution, result delivery, and cursor lifetime, alongside the upgrade to Arrow 25. v1.34.0
Less wasted work: the server interrupts queries when their clients disconnect. v1.38.0
A dedicated Metabase driver: our first GizmoSQL-specific release builds on the community Flight SQL driver and uses GizmoData’s JDBC driver. Metabase v1.0.0
JDBC improvements: corrected timezone handling and refreshed upstream dependencies. Version 1.7.0 now requires Java 17. JDBC release notes
🌊 Community spotlight: 18.7 billion maritime records
Aalborg University’s Seagull platform makes ship movement data available through interactive maps: more than 18.7 billion AIS records from 126,400 ships.
The team wanted DuckDB’s performance while allowing Airflow, its tile server, and the web application to share the database. GizmoSQL provided that shared access over Arrow Flight SQL.
Tiles are computed on demand, with previously served tiles cached by Martin. Kasper F. Pedersen’s guest article explains how the pieces fit together. Read the Seagull story.
👋 Welcome, Chris—and come join the conversation
Chris Wones has joined GizmoData as Strategic Advisor. His 34-year career spans James River, Retek, dunnhumby, and Kroger/84.51°. Chris and Philip worked within the same enterprise data organizations for years; he brings extensive experience evaluating and operating the platforms GizmoData is helping teams rethink. Read the announcement.
We also launched a GizmoData community Slack in June, and Philip joined Torben Andersen on the A/I Prosperity podcast to discuss GizmoSQL and warehouse costs. You’ll find the invitation and episode on our news page.
Building something with GizmoSQL? Reply and tell us about it. We’d love to feature more projects, hear what’s working, and learn what you need next.
Thanks for sticking with us through the newsletter’s summer vacation. The ducks have been busy. 🦆