All about Databasin One
What it is, what it does, and why it's different.
Databasin One is an AI agent that works across all the data you've connected. You ask it something in plain English; it figures out which data to look at, runs the queries, analyzes the results, and hands you back an answer — with charts, with tables, with PDFs if you want them.
It's not a chatbot bolted onto your warehouse. It's an agent that picks its own path through your data, budgets its own work, and shows you exactly what it did.
Open Databasin One from a project Meet the agent See the skills
The headline capabilities
| What it can do | Example question |
|---|---|
| Answer factual questions against your data | "How many active Salesforce opportunities do we have by region?" |
| Make a chart on demand | "Chart that by week." |
| Refine or swap the chart | "Make it a stacked bar, colored by segment." |
| Generate a full report as a PDF | "Write me a one-page summary of Q1 pipeline health." |
| Search inside uploaded documents | "What did the contract say about termination notice?" |
| Build a curated semantic model | "Model these tables so I can ask about revenue by customer." |
| Build a whole dashboard from questions | "Build me a sales dashboard that answers these five questions." |
| Keep working in the background | "Do a full quarterly review — I'll check back in 10 minutes." |
How it's different from a plain LLM with SQL access
Three things, mainly.
It knows what it's allowed to do
Databasin One has 12 tools: 7 skills — reference knowledge it reads on demand — and 5 data/action tools that actually touch your warehouse (get_schema, get_sample_data, execute_sql, and two for building semantic models). It can call them in any order, but it can't do anything you didn't grant it. Your data stays in your warehouse; the agent just orchestrates.
It budgets itself
Every conversation has a query budget — by default, three execute_sql calls max. The agent plans its path knowing that constraint, so it doesn't burn your credits brute-forcing. Schema reads, sample-data peeks, and skill calls are free; only real SQL counts.
It shows its work
Every answer comes with:
- The narrative (the explanation).
- The data (the rows used to build the answer).
- The sources (which tables, which queries, which uploaded documents).
You can click any source to see the exact SQL that produced it.
Seeing the SQL the agent ran is the difference between trusting an answer and hoping it's right. Every time.
It runs on your choice of model
Databasin One isn't tied to one model. It runs on Claude or GPT, depending on which LLM connector you point it at — the same agent loop, the same tools, the same query budget either way. The two behave a little differently in the details, and picking one is its own short read: Choosing your model.
The two modes
Databasin One runs in Agent mode by default — the mode that plans queries, runs them, and shows its work. There's also an AI Assistant mode you can flip to from the same toggle, for a more conventional back-and-forth chat. Most people stay in Agent mode.
The things you do most
1. Ask a question
Type. Hit enter. Watch it work.
2. Chart an answer
Every data answer has a "Chart this" action. Databasin One proposes a couple of sensible chart options; you pick one. You can quick-swap between bar, line, pie, and scatter, or ask for a refinement in words ("make it log scale").
3. Add it to a dashboard
Any chart has an "Add to dashboard" action. It drops the tile onto the Dashboard Canvas, where you can arrange tiles and publish to your Gallery for the team.
4. Generate a report
Ask for "write me a summary" or "generate a PDF about Q1 revenue." If the data is already in the chat, it fast-paths and stays inline; a from-scratch report runs as a background task so you can keep working.
Newer surfaces worth knowing
Databasin One has grown well past chat-and-charts. A few things you might not have found yet:
- Choosing your model — run the agent on Claude or GPT.
- Build a dashboard from questions — the AI Dashboard Builder turns a list of questions into a cohesive dashboard.
- Discover — let the agent surface what's interesting in your data.
- Build a semantic model by chatting — turn raw tables into a curated model through conversation.
- Bring Power BI models into One — upload a Power BI model and ask it questions in plain English.
- Chat with a dashboard's data — let viewers question the data behind a published dashboard.
Where things live in the UI
- Main chat surface — center of the screen.
- Data context — left sidebar. Pick which warehouses, tables, and documents the agent can see.
- Downloads — slide-in panel on the right, showing in-flight and completed documents.
- Workspaces — name a session and come back to it later (see Workspaces).