The GizmoData Dispatch - October 2026 Edition

The GizmoData Dispatch - October 2026 Edition

🐂 A new server, a public cloud, and a whole ecosystem that showed up on day one.

GizmoData

Oct 28

GizmoData Cloud opened to everyone on September 28, and October was the busiest month in GizmoData’s history: we launched Hardcharger™, a new open-source server for DuckDB 2.0, and its drivers and integrations shipped alongside it. Meanwhile, GizmoData Cloud gained Hardcharger clusters, a full SQL editor and new ways to keep costs down.

Let’s get into it. 👇

🐂 Meet Hardcharger: the DuckDB 2.0 server

On October 21 we launched Hardcharger, the successor to GizmoSQL. It’s written from scratch on DuckDB 2.0, and it’s free: Apache-2.0, with no editions and no license keys. Everything that was an Enterprise feature in GizmoSQL is in every Hardcharger build.

Why a new product instead of a GizmoSQL 2.0? Running GizmoSQL in production with customers taught us three things:

• The transport was the friction. Arrow Flight SQL runs on gRPC: HTTP/2 end to end, gRPC-aware proxies, generated code in every client, and no plain browser client.

• Security belongs in the core, not left for each server to get right on its own.

• Features shouldn’t be gated. Every team running a shared DuckDB needs access control, SSO and instrumentation.

DuckDB 2.0 was the moment to fix all three at once. Read the launch post.

⚡ Warp SQL 1.0: SQL over HTTP, results as Arrow

Hardcharger speaks Warp SQL, a new open protocol (Apache-2.0) that we wrote. It keeps the best part of Flight SQL, Apache Arrow on the wire, and replaces the rest with ordinary HTTP and TLS:

• Short queries take one round trip. One POST runs a statement and streams its rows back.

• Long queries survive real networks. A dropped download resumes from the batch where it stopped.

• Big results and uploads go wide, over several connections at once; a multi-part bulk load commits as one transaction.

• It works with what you already run. Any proxy, load balancer or API gateway that passes HTTP passes Warp SQL, and so do curl and a browser’s fetch.

Version 1.0 is final: from here on it changes only by addition. It has an OpenAPI 3.1 description and a conformance suite of 282 tests. The spec is on GitHub.

🔐 Security in the core: RBAC and the rest, all included

This is where Hardcharger differs most from the usual “DuckDB behind a port”:

• Role-based access control checked against each statement’s own plan, inside DuckDB, so nested SQL, views and macros can’t route around a grant. Anything the check can’t classify is refused.

• Row policies, column masking and tenants, plus a tamper-evident audit log.

• Single sign-on with your OpenID Connect provider, on the server’s own port, with users optionally created on their first sign-in.

• Mutual TLS with certificates as identity: tokens from a certificate sign-in are bound to that certificate, so a copied token is useless without the private key.

• Every session, statement and result belongs to its owner and is checked on every use.

• The host’s credential files (SSH keys, cloud credentials, the server’s own TLS key) are refused at DuckDB’s file-system layer, whoever asks, the administrator included.

It’s also built to be watched: instrumentation into DuckDB or PostgreSQL, a Prometheus /metrics endpoint, query profiles and KILL SESSION.

🖥️ A web UI in the binary, a desktop app, and an app for your phone

Start the server with --web-ui and you get a SQL workspace on the server’s own port: up to 20 editor tabs, each with its own session; a results grid for up to a million rows that copies cleanly into Excel; a schema browser that follows access control; and uploads of CSV, Excel, Parquet, JSON or Arrow files straight into tables. Version 2.1 added a SQL Monitor (live statements, queue status, query profiles) and a Server Configuration page.

Hardcharger Desktop arrived with v2.1.0 on October 26, and it’s a complete Hardcharger server in one download. Inside is the same hardcharger-server you’d run in production, the hardcharger command-line client, and a native app for macOS 15+ on Apple silicon and for Windows that starts and stops the server, opens the SQL workspace and follows the logs. There’s nothing else to install. Close the window and the server keeps running in the menu bar or system tray, so your notebooks, BI tools and drivers can connect to it like any other Hardcharger server, and you can share it on your network when you want to. Download it. Hardcharger for iPhone and iPad runs the whole server on the device, and it’s coming soon.

⏱️ Benchmark corner: 1 TB in 42 seconds, for about 9 cents

On one AWS Graviton5 machine (r9gd.24xlarge), Hardcharger 2.0.0 ran 22 queries derived from TPC-H over 1 TB in 42.4 seconds, about $0.09 of compute at the on-demand price. The same machine ran our earlier GizmoSQL 1.42.0 baseline (DuckDB 1.5.6) in 77.0 seconds, so that’s 1.8× faster.

To be fair about where that speed comes from: most of the query-time gain is DuckDB 2.0 itself. Where Warp SQL pulls ahead is moving data across a network. Between two AWS machines with TLS on, fetching 60 million lineitem rows (10.5 GB of Arrow) took 0.97 seconds against GizmoSQL’s 15.2, and bulk-loading the same table was 3.1× faster. Methods and full results.

Derived from the TPC-H Benchmark; not an official TPC-H result and not comparable to published TPC-H results. TPC-H is a trademark of the Transaction Processing Performance Council.

Try it in about a minute:

curl https://install.hardcharger.gizmodata.com | sh

Or read the documentation.

🔌 An ecosystem on day one

The drivers and integrations launched with the server, not months after it:

• Python and Go (ADBC): pip install adbc-driver-hardcharger

• JDBC on Maven Central, and JavaScript/TypeScript on npm (@gizmodata/hardcharger)

• ODBC for Windows (a signed MSI), macOS and Linux

• Hardcharger MCP server and Claude Desktop extension, so your AI assistant can query it

• dbt, SQLMesh and SQLFrame adapters

• adbc-replicate-hardcharger for copying Snowflake, Oracle, PostgreSQL or DuckDB schemas in (more below)

• A signed Power BI connector, a QGIS plugin, DBeaver setup, a PDF Loader, and querying Hardcharger from DuckDB itself through ADBC

See every integration.

🦆 What about GizmoSQL?

GizmoSQL keeps its name and its Flight SQL protocol, and we’ll maintain it for about six months. It shipped three releases this month: v1.41.0 introduced an experimental Edge channel on DuckDB 2.0, v1.42.0 moved to Apache Arrow 26, and v1.42.2 put Edge on the final DuckDB 2.0.0 release. For new projects, we recommend Hardcharger. GizmoSQL releases.

☁️ GizmoData Cloud is open to everyone

In September we told you hosted GizmoData Cloud was in early access. Since September 28 it’s generally available: managed DuckDB clusters, DuckLake catalogs and federated queries on AWS, Google Cloud and Azure, from one console. Read the announcement.

Sign up and you’ll get a $25 launch credit, applied automatically, with no credit card needed. Start free. Paid plans start at $29 a month.

Here’s what changed in the month since launch:

• Hardcharger clusters are the default for new clusters (GizmoSQL is still available), with an Access Control page for its roles and grants.

• A real SQL editor: up to 20 tabs, each its own session, with transactions that span runs; a result tab per statement; autocomplete; saved queries in folders; a data explorer; and Cancel.

• Paying only for what runs: Auto-Stop after a set time, and as of today, stop when idle, which stops a cluster once no statement has run for its idle window. Auto-Stop and Auto-Delete can both be changed after a cluster is created.

• Endpoints on your terms: the shared gateway is included; a dedicated load balancer for long-lived connections starts at $0.02 an hour.

• More catalogs: bring your own Iceberg REST catalog, test any catalog connection before saving it, and see what the nightly DuckLake maintenance did on each catalog’s page.

• Self-serve: API keys, Slack alerts for your organization, Start Again for a failed cluster, autoscaling you can switch on or off later, Stable/LTS/Edge release channels, a Help panel on every page, and a guided tour.

Want to see the three clouds side by side? We raced six 64-vCPU clusters on AWS, Google Cloud and Azure over 1 TB: AWS Graviton5 ran the fastest lap (79 seconds for all 22 queries, $0.30), Google Cloud took four of six golds, and DuckDB 2.0 beat 1.5.6 in every lane. The whole race took 38 minutes and cost about $37. Read “Three Clouds, One API”.

🔁 Replicate once, query all day

adbc-replicate-hardcharger launched with Hardcharger and copies whole schemas from Snowflake, Oracle, PostgreSQL, DuckDB or another Hardcharger into Hardcharger. You describe what to copy in a YAML file: schemas, tables and views by pattern, and the columns to leave out or mask. Data streams as Arrow from end to end over Warp SQL’s bulk ingest, masks run inside the source, views are translated, and each run copies only what changed. It’s on PyPI (pip install adbc-replicate-hardcharger), with a container image and a Helm chart for running it on a schedule.

It grew out of adbc-replicate, the GizmoSQL tool we introduced on October 16. Two customers, one on Snowflake and one on Oracle, already replicate every hour into GizmoData Cloud, and their AI agents query the copy over MCP all day without touching the source. Read that story.

🤝 Customers and case studies

We published our first two case studies:

• Vaden Automotive replicates its Snowflake data into GizmoSQL inside its own AWS account. Employees ask business questions in plain English through Claude, and internal apps show real-time data. “It’s fast, lightweight, and delivers true business intelligence across our entire dealership group without the large price tag,” says Chris Chilton, Vaden’s CIO.

• DESS at Aalborg University: the Seagull maritime platform we featured last month has grown to 23.8 billion AIS records from more than 235,000 ships, with one DuckDB database shared by its data pipelines and live maps.

Truck Cab Manufacturers (heavy-duty truck cab manufacturing, on GizmoData Cloud) and Company Wrench (heavy equipment and attachment rental and sales, on GizmoData Cloud at Customer) also joined our customers page.

📰 Daniel Beach tried to break it

Daniel Beach of Data Engineering Central took GizmoSQL and GizmoData Cloud for a spin in “DuckDB + ArrowFlight = GizmoData”. He ran GizmoSQL in Docker with Python, dbt and Airflow, then hit it with eight concurrent writers while readers polled. His verdict: “I’m a little disappointed I couldn’t break it.” After that he ran a three-year window-function workload over Iceberg tables on S3 on a small GizmoData Cloud cluster. His summary of the idea: “Take DuckDB, wrap it in Arrow Flight SQL, harden it into a real client-server setup, and hand it to me. What more could a data guy ask for!?”

His wish list includes Iceberg catalogs like Polaris, and GizmoData Cloud now attaches Iceberg REST catalogs, Polaris among them. Thanks, Daniel!

🇧🇷 Olá, Brasil

GizmoData now has a full Brazilian Portuguese site at gizmodata.com.br, with the Hardcharger launch post, product pages and Cloud pages translated and checked against the English.

👋 Come build with us

If you run DuckDB anywhere other than a laptop, give Hardcharger a try and tell us what breaks. Questions, ideas and war stories are welcome in the GizmoData community Slack, or just reply to this email.

Thanks for reading, and see you in November. 🐂🦆

Thanks for reading The GizmoData Dispatch! Subscribe for free to receive new posts and support our work.

Subscribed

The GizmoData Dispatch is designed to highlight our products and services as they relate to the data analytics landscape. We appreciate your readership and support!

© 2026 GizmoData LLC
6915 Ponce De Leon Blvd, North Port, Florida 34291