auto.selectLive dashboards from the AI assistant you already use

Make more of agood question.

Connect your databases and data lake, then ask your AI assistant what you want to know. The answer comes back as a live dashboard your team can open, filter and share.

Join the waitlistSee how it works

The idea

An answer in a chat is gone a few scrolls later. auto.select keeps the good ones. A question becomes a dashboard you come back to, a workspace your team shares and a monitor that keeps watching.

01

Ask.

Start with something you’ve been wondering about. Your assistant looks through the tables you’ve connected, tries a few queries and finds the answer.

02

Answer.

The answer arrives in the conversation as a dashboard. Open it and it stands on its own, with live numbers you can check again tomorrow.

03

Refine.

Ask a follow-up and the same dashboard gains a filter or a breakdown, so the answer keeps improving in one place.

04

Gather.

Share a dashboard by its address. Gather related ones into a workspace, where everyone’s questions sit under the same filters.

Try the shared filter

05

Watch.

Leave a monitor on the number that matters. When its condition is met, it records an observation: a milestone, an opportunity or a change worth a look.

Illustration · fictional data

Under the hood

Every answer is a short script.

  1. 01

    Your assistant writes it.

    auto.select keeps the script and reruns its queries whenever someone opens the dashboard. Ask for a change and your assistant edits the script.

  2. 02

    It learns your data.

    When your assistant finds something the schema can’t say, like trials logged as upgrades, it leaves a note on the data. Whoever asks next starts from that note.

  3. 03

    Your team decides what it can see.

    Rules for each team decide which tables and columns its queries can reach. A hidden column can’t be named, whether a person or an assistant writes the SQL.

  4. 04

    Old dashboards keep working.

    Each script names the edition of the language it was written in, and keeps running on that edition as auto.select changes.

growth.starIllustrative script
1# autoselect: 123region = binding.one(4    id = "region",5    label = "Region",6    options = ["Europe", "Americas", "Asia Pacific"],7)89upgrades = sql(10    """11    SELECT week, count(*) AS upgrades12    FROM upgrades13    WHERE region = $region14      AND NOT is_trial  -- trials are logged as upgrades15    GROUP BY week16    """,17    {"region": region},18)1920dashboard(21    title = "Growth",22    body = layout.grid(23        template = "headline trend",24        areas = {25            "headline": viz.big_number(26                value = upgrades, calc = "sum", column = "upgrades",27            ),28            "trend": viz.bar(query = upgrades, x = "week", y = "upgrades"),29        },30    ),31)

Where it runs

Queries run where you choose.

A gateway runs each query and sends back what the chart needs: twelve weekly totals rather than two million rows. Run your own beside your databases and the counting happens next to the data. Or let auto.select run it, with nothing to install.

Setup
Nothing to install
A Linux machine beside your databases
Your databases
Reachable from the internet, so auto.select can connect in
Stay behind your firewall. The gateway dials out over an encrypted tunnel.
Compute
Ours
Your machine’s

Questions

Which AI assistants does it work with?
Claude, on the web and in the desktop app, through a custom connector. auto.select speaks MCP, the open standard for assistant connectors, so other assistants that support it can connect too.
Which databases can I connect?
PostgreSQL, MySQL, SQLite, DuckDB and MotherDuck today. A data lake connects through your own DuckDB server or MotherDuck, with its tables attached there. Tell us what you use when you join the waitlist; it decides what we support next.
Does the AI see my data?
Yes. Your assistant reads table structures and sample rows while it works out a query, and a dashboard’s numbers reach whoever opens it. Each team’s rules decide which tables and columns its queries can reach. Connect a read-only user, and an analytics view if you have one.
When can I start?
We’re inviting people a few at a time. Join the waitlist and we’ll email you an invitation when there’s a place.

What have you been wondering about?

auto.select is in early access. Join the waitlist and we’ll email you when there’s a place for you. We only use your address for that.

Which databases do you use?Optional. It helps us decide what to support next.
  • PostgreSQL
  • MySQL or MariaDB
  • SQL Server
  • SQLite
  • DuckDB or MotherDuck
  • BigQuery
  • Snowflake
  • Databricks
  • Redshift
  • ClickHouse
  • Parquet, Iceberg or Delta files
  • Something else