briqbriq
Dashboard
All postsNews

A shared report in 30 seconds, with DuckDB

Ask your agent for a report on a dataset. It loads the file into DuckDB, writes the dashboard in SQL and sends you a link.

Sylvain Goumy

Founder, briq

20 October 2026 · 2 min read

Ask an agent for a report on a dataset and you usually get a table in the chat. It read the rows into its context to build it: slow, expensive, and nothing you can send to anyone.

We want a link to a dashboard instead. And we want it before you have switched tabs.

DuckDB does the work

DuckDB is an analytics database that runs inside the program that uses it. There is no server to start and no warehouse beside it. It reads CSV, Parquet and JSON straight from a URL, and it is built for group-bys over millions of rows.

For the dashboards we use Shaper, an open source tool built on DuckDB. A dashboard is a SQL file:

SELECT 'World generation by source (TWh)'::LABEL;
SELECT make_date(year::INT, 1, 1)::XAXIS, source::CATEGORY, twh::LINECHART FROM world;

Each SELECT is a chart, and its casts say how to draw it. Agents write SQL well, so they write dashboards well.

What the agent does

You ask:

Start DuckDB on briq, load 'Our World in Data' energy dataset into it, generate a report on world energy generation, and share it with me.

The agent:

  1. Starts the duckdb recipe: one briq, with DuckDB and Shaper set up.
  2. Loads the file: one SQL statement. DuckDB downloads it itself, so the data never goes through the agent.
  3. Writes the dashboard: one SQL file, two charts.
  4. Shares it: a read-only dashboard behind the briq's private link. Only people with the link see it.

What we measured

We ran the same task on briq, again and again, and timed it from the outside: a headless browser opens the shared link and waits until both charts are drawn.

  • Start: 12 to 15 seconds for the briq to boot, with DuckDB and Shaper ready.
  • Load: under 5 seconds for the 23,000-row, 130-column energy file.
  • Dashboard: under a second to save it, about 2 seconds to draw it.
  • In all: 21 to 24 seconds from the agent's first call to both charts on screen.
  • Later: the briq sleeps when nobody uses it. The next visit wakes it, and the charts are back in about 13 seconds. The visitor after that waits under a second.

It all runs on our smallest size. The agent can also save the dashboard as a PDF, to attach to a message.

Compared to Metabase

Our first reporting recipe ran Metabase on a Postgres warehouse, and it is still in the cookbook for people who build dashboards by clicking. On the same task it took two and a half to almost eight minutes: a large Java application boots slowly on a small machine.

From the agent's first call to two charts drawn behind a shared link. One brick is ten seconds.

Try it

1

Add briq to Claude

In Claude: Settings → Connectors → Add custom connector.Name it briq-stg, paste this URL, then click Connect.

https://api.stg.briq.cloud/mcp

In Claude: Settings → Connectors → Add custom connector.Name it briq-local, paste this URL, then click Connect.

http://localhost:3001/mcp
2

Connect Claude to briq

Sign in, or create your account and claim your €10 of free credits.

3

Try it

Deploy briq-cloud/wall github repo on briq (https://briq.cloud/docs/guides/run-a-github-repo.md) and give me a link to it.
Deploy briq-cloud/wall github repo on briq (https://stg.briq.cloud/docs/guides/run-a-github-repo.md) and give me a link to it.
Deploy briq-cloud/wall github repo on briq (http://localhost:3000/docs/guides/run-a-github-repo.md) and give me a link to it.
Manual installations

The full walk-through is the guide Report on a dataset. Tell us what you report on: hello@briq.cloud.

Sylvain Goumy

Founder, briq

LinkedInXGitHub

How can we help?

Message to the briq team
hello@briq.cloud