Analysis
Asking questions
Ask the agent a question in a sentence, in any language, and get back SQL, a chart, and the assumptions it made.
The Agent page answers questions about your data. Ask in a sentence, such as "Which lead sources produced the most closed-won revenue this year?", and Datachef writes the SQL, runs it, and charts the result.
Ask in your own language
Ask in whichever language you work in, and the answer comes back in it too:
¿Cuál es el valor de las oportunidades de venta abiertas que se consideran de calidad?
Datachef reads the question, writes the SQL against your tables, and explains the result and its assumptions in Spanish. If your workspace has a qualified_pipeline model, it's used here too, whatever language the question is in.
What an answer includes
- A title and the result, with a chart chosen to fit it.
- The SQL it ran, which you can open and edit in the SQL editor.
- An explanation of what the query does.
- The assumptions it made, such as which stage counts as closed-won, or which date it filtered on. Read these first. They're where a plausible-looking answer can quietly differ from what you meant.
If the SQL fails, the agent tries to fix it once before reporting the error.
It uses your models
The agent is given your workspace's models along with its tables. If you've saved a qualified_pipeline model, a question about qualified pipeline is answered with your definition, not one the agent made up. The best way to get consistent answers is to save the definitions your team agrees on as models.
Long-running questions
A question runs in the background for up to five minutes. While it runs:
- The question is locked. Choose Stop to cancel it and edit the question.
- Leaving the page doesn't stop it. The answer is saved on the tab, so reopen the tab to see it.
What's sent to the AI model
Only your question and your workspace's schema (table names, column names and types, and model definitions) are sent to the AI model provider. Your rows aren't sent. The query runs inside Datachef, against your synced data.
Usage allowance
Each organization has a monthly allowance of AI tokens, which depends on your plan. Owners and admins can see usage under Settings → Usage. When the allowance runs out, the agent stops answering until the next month. Querying in the SQL editor isn't affected. Contact us to raise the limit.
Tips
- Name the systems and the time range. "Closed-won Salesforce opportunities in Q3, by Marketo lead source" beats "how did marketing do?".
- Ask for the definition you want. "Count a customer as active if they had a usage event in the last 30 days." Then save the answer as a model so nobody has to say it again.
- Check the assumptions before you share the number.
Something missing or wrong on this page? Tell us