Chat page
The Chat page is where you ask questions about your connected data, or general questions. Each thread is attached to one workspace and one connection.
What the Chat page is for
Chat lets you ask business or operational questions in natural language. For a question about your data, STYD translates it into a controlled database query, validates it, executes it safely, and returns a structured answer. A general question, a request for advice or a draft is answered from general knowledge and labelled General knowledge.
Chat layout
Create, select, and switch between chat threads.
Ask questions and review answers.
Shows which database connection the active thread is using.
Type your question and send it to STYD.
Create a chat thread
Connection badge
At the top-right of the chat area, STYD shows the connection name used by the current thread.
Confirms which database the thread is attached to.
Appears when AI Guide is still running or when schema analysis may be incomplete.
Status line while it runs
While STYD works, one status line under your question shows the current stage, with a placeholder where the answer will appear. If an earlier answer in the thread is still running, it says “Waiting for current answer to finish”.
What STYD is doing now, for example “Understanding your question” or “Preparing query and running query”. The stages are listed below.
The line is replaced by the answer when it is ready.
“This answer could not be finished. Try asking again.” The technical reason stays out of the way and is in “How I found this”.
Status stages
During execution, STYD may show more detailed status labels. These labels make the pipeline easier to understand.
STYD is initializing the request and saving the user message.
STYD is identifying the intent, entities, filters, and expected result.
STYD decides how to answer the question and which data path to use.
STYD searches for the most relevant tables.
STYD reads table and column structure from the AI Guide.
STYD identifies the columns needed for filters, grouping, metrics, or joins.
STYD prepares the SQL query.
The validated query is executed against the database.
STYD checks whether the result matches the expected answer shape.
STYD checks whether generated claims are supported by the data.
STYD formats the final response as text or a table.
The answer from your data is already shown; STYD is writing the short General knowledge part that follows it.
The answer is saved to the thread history.
Result tags under an answer
Some answers include small tags below the assistant response. These tags help you understand where the answer came from and how confident the system is.
The answer was generated from database results.
The answer may rely on model reasoning instead of direct database output.
The answer was written from general knowledge, not from your data.
Indicates how confident STYD is in the generated answer.
Shows whether answer claims were supported by available evidence.
How I found this
When available, STYD shows a link called How I found this under the AI response. This opens a trust panel that explains, in plain language, how the answer was produced from your data. For an answer from general knowledge, it says: “Written from general knowledge and your business context, not from your data.”
What the panel includes
- Simple explanation, a short summary of what STYD did.
- Business logic used, the calculation or selection logic behind the answer.
- Filters applied, the conditions used, such as company, invoice type, payment status, date range, or customer filters.
- Data used, the business objects or tables used to answer the question.
- Assumptions, important interpretation notes, such as when no date range was requested.
- Technical details, on a database, the SQL query and technical trace; on Odoo, the Odoo records and filters read. Collapsed by default for non-technical users.
Example
If you ask for draft invoices, STYD may explain that it searched customer invoices, kept only invoices with draft status, applied your accessible Odoo company scope, and returned the matching records.
This panel is designed to help business users trust the answer without needing to read SQL. On PostgreSQL and MySQL, the SQL remains available under technical details for advanced users, debugging, or audits.
Feedback buttons
Each assistant answer can be reviewed with thumbs up or thumbs down feedback.
Mark the answer as helpful. This can help similar answers rank better later.
Open a feedback form to explain what went wrong.
Thumbs down feedback
When you mark an answer “Not helpful”, a short form asks what went wrong. You pick the closest reason. There is no field for SQL or notes.
Wrong number, wrong period, wrong customer, product or filter, missing data, STYD did not understand my question, wrong language, too slow, or other. On a database connection there is also “wrong data source”. Answers taken from your uploaded files have reasons of their own, such as wrong document or wrong citation.
An answer from general knowledge offers “Wrong or unhelpful advice”, “Too generic”, “Should have used my data” and “Out of date”, plus misunderstood question, wrong language, too slow and other. An answer from your data that ends with a General knowledge part also offers “Wrong or unhelpful advice”.
One optional field where you can describe the answer you expected.
Parallel prompts and sequential answers
You can send multiple prompts one after another. STYD keeps the thread context and returns answers in order so the conversation remains consistent.
Best practice
Start with a clear question, review the result, then refine with follow-up questions instead of starting over.