an agent, live
Ask in English. The agent writes the spec. 0sql writes the SQL.
The agent from the chat dashboard demo, on the same customer-service model, driven by GPT-5.4 mini. It never writes SQL: it searches the model for fields, writes a query spec, and 0sql plans the statement with the caller's row-level security compiled in. Every turn, tool call and token is in the trace. Nothing is run: this page shows the SQL, not the rows.
The agent and the model are open source. Read how it is built, or clone it and run it on its own warehouse.
Rather call the API yourself? Write the request line below
Ask about contact volume, handle time, resolution or cost, or start with one of these.
thinking 0.0s
the query API, live
Call the API. Read the SQL.
Submit your own inputs against a real model. See how easy it is to construct a query from your apps and get back reliable, secure SQL. Pick a model, pick who is asking, take a preset or write your own, and watch joins, grain and row-level security get settled, or a refusal name what could not be resolved.
{
"expr": "month(date), region, contact rate, date > 28d",
"context": {
"email": "dana@example.com",
"groups": [
{
"name": "CS Ops AMER",
"tags": [
"region:AMER"
]
}
]
}
}Every statement is planned by app.0sql.io as you ask, through a proxy that holds a read-only key for these two projects. Nothing executes. For more lines, read the chat dashboard walkthrough.
- send
- read
- resolve
- plan
- SQL
-- the statement arrives hereContact Rate is a compound across two facts: answered contacts over the membership base. The base is a snapshot measure, so the planner finds the last snapshot in each month first, then divides. Three CTEs, and nobody wrote them.