Limitations & philosophy

STYD is built to make database answers easier to access, safer to generate, and simpler to verify — but it is not designed to hide uncertainty.

Our philosophy

STYD is designed around a simple principle: answers should be useful, but also explainable.

The goal is not to make AI look magical. The goal is to make data access faster, safer, and easier to trust.

This is why, for questions about your data, STYD focuses on the AI Guide, SQL validation, read-only execution, and transparent answer generation instead of blindly sending questions to a model. A question that needs no data, such as a definition or advice, is answered from general knowledge and labelled General knowledge.

What STYD does well

Answering questions from structured database data
Helping users explore tables, metrics, and records
Generating SQL with validation and guardrails
Explaining how an answer was produced
Answering general questions and writing drafts, labelled General knowledge

Current limitations

STYD can still make mistakes, especially when the database structure is ambiguous, business definitions are unclear, or the requested metric requires context that is not available in the schema.

Ambiguous business terms
Words like “revenue”, “active”, or “churn” may need a clear definition.
Incomplete schema context
If table or column names are unclear, the AI Guide becomes more important.
Permission boundaries
Answers from your data use only the data and tables available to the current workspace or user.
General answers
They come from general knowledge, not from your data, and may be out of date. On legal, tax, medical or investment topics they give general principles only.
Critical decisions
Important financial, legal, or operational decisions should always be verified.

What STYD does not do

Other sources
STYD reads PostgreSQL, MySQL and Odoo (15 and later), plus the documents and spreadsheets you upload. Other kinds of database are not supported.
Writing to your data
Never. STYD only reads: one SELECT per question on a database, read-only reads on Odoo.
Sending
Never. STYD can write a draft email or message, but you copy it and send it yourself.
Long or slow queries
A query stops after 15 seconds by default, and a result larger than 5 MB is refused. Long lists are cut: the answer says how many rows it shows.
Secrets
Passwords, tokens, keys, and card or bank numbers are always shown as [hidden], and are never sent to the AI model.
Private networks
A database on localhost or a private network cannot be reached from STYD Cloud, and there is no SSH tunnel option.
Viewers
A viewer can read conversations but cannot ask questions.
Odoo questions that are refused
A question STYD refuses as unsupported still counts as an answered question and uses one prompt. Errors and setup steps do not.

What is sent to the AI provider to answer a question is described in the Privacy Policy.

How to get better results

The quality of answers improves when your schema is documented, your business definitions are clear, and users ask specific questions.

Analyse the database again after adding new columns (it costs prompts)
Use clear date ranges and filters
Review how important answers were calculated
Define important business terms

Trust, but verify

STYD is designed to help users reach correct answers faster, not to remove the need for judgment. For important results, use the calculation details (the SQL on a database), the trace, and feedback tools to review how the answer was generated.