Go from question to analysis in minutes.
An agent with schema-aware access to your sources, working in a sandboxed cloud notebook your team can rerun and share — with credentials that never leave the broker.
revenue-by-week.ipynb
python 3 ·idle
It's still happening.
Your numbers live in Postgres, a warehouse, and three SaaS tools.
Answering one question means making sure all the creds are in place on your local machine, prompting an agent that has never seen your schema, and sending screenshots or sharing a notebook that someone else has to boot up.
what that costs
Most of the afternoon goes to exports and column names. The question gets what's left.
Three things that fix it
Each one answers a step in the loop above — credentials, schema, sharing.
Credentials never leave the broker
Connect sources with OAuth once. The extractor holds warehouse credentials and the gateway holds the model key — neither secret lands on a laptop or enters the sandbox.
Your agent already knows your schema
Attach context per source, table, or column — the abandoned field, the timezone, the join that looks wrong. The agent reads it when it writes the extract, not after guessing at column names.
Send a link, not a screenshot
Notebooks run in the cloud against your connected sources. Rerun last quarter against this quarter, or hand a teammate the URL instead of a file they have to boot locally.
How it works
01
Connect a source
OAuth into Postgres, hit Test, and read back the catalog we discovered. Nothing is copied anywhere yet.
02
Leave notes on the schema
Write down the quirks a new hire would need — the abandoned column, the timezone, the two tables that disagree. The agent reads them the same way.
03
Ask for the analysis
Describe the question and the shape of the answer. The agent previews tables, runs an extract, writes cells, and executes them.
04
Interrogate it, then share it
Edit any cell and rerun it yourself. Close the tab and the sandbox pauses with kernel state intact; reopen tomorrow and it resumes.
Not every question is an aggregation.
A dashboard lets you explore whatever someone has already modeled, and it does that well. The wall is the question the model can't be asked — a cohort defined three ways, a fuzzy join across two exports, a pass over free-text feedback. That is a notebook's grammar rather than a chart's, and here you get it without modeling the data first.
what a dashboard is good at
- The number you already track, on a schedule
- Drilling into a model somebody already built
- One definition of revenue, enforced everywhere
what this is good at
- The question that needs a column nobody modeled
- Anything you'd rather write in Python than draw
- An answer you can rerun next quarter and send as a link
What we publish
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.
Before you ask
- Does our data leave our infrastructure?
- Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
- What happens when the schema changes underneath us?
- Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.
The vantage point, without the exposure.
A sandboxed notebook and an agent for every analysis, reaching your sources through a broker rather than a copied connection string.