Query plan & execution flow
When you ask a question, STYD does not just guess. It works through a series of steps to turn your question into a safe, read-only query and a checked answer.
The flow, step by step
1. Your question is received
It is added to the thread so the conversation has context.
It is added to the thread so the conversation has context.
2. STYD understands the question
It works out the intent, the entities, the filters, and the kind of answer expected.
It works out the intent, the entities, the filters, and the kind of answer expected.
3. It plans an approach
STYD decides how to answer and which data path to use. A question that needs no data, such as a definition, is answered from general knowledge instead, labelled General knowledge, and skips the steps below.
STYD decides how to answer and which data path to use. A question that needs no data, such as a definition, is answered from general knowledge instead, labelled General knowledge, and skips the steps below.
4. It uses workspace metadata / AI Guide
It selects the relevant tables, reads the schema, and finds the needed columns.
It selects the relevant tables, reads the schema, and finds the needed columns.
5. It prepares a query plan
STYD builds a controlled, read-only query.
STYD builds a controlled, read-only query.
6. It runs the query
The query runs read-only against your data, within your access.
The query runs read-only against your data, within your access.
7. It validates and verifies
STYD checks the result shape and whether the answer’s claims are supported by the data.
STYD checks the result shape and whether the answer’s claims are supported by the data.
8. It formats the answer
The result is presented as text, a metric, a table, or a chart.
The result is presented as text, a metric, a table, or a chart.
9. It saves to the thread
The answer is stored in the thread history.
The answer is stored in the thread history.
Faster starts from cache
STYD keeps your database layout and the context it needs in memory for a few minutes, so the next question starts faster.
When it cannot answer directly
If a question is ambiguous, the data is restricted, or the request is unsupported, STYD will ask for clarification or return a clear limitation rather than guessing. When your data does not hold what you asked, it says so and can add general guidance labelled General knowledge, never invented figures. This is by design, it is safer to flag uncertainty than to invent an answer.