Troubleshooting the SQL chat

General tips

• The SQL chat is fully supported with GPT-5.4, so if you are using another model, support is expected to be limited at the moment and thus we recommend to switch to GPT-5.4 using the chat options.

• If it is not clear which questions to form and how, consider asking amber "Which databases, tables and columns do you have access to? I want to know what I can retrieve with the sql connector."

Expected behavior

• amber is expected to perform retries in order to create the correct SQL query for your input question and database schema.

• If a query is a complex analytics report that touches many concepts or a broad time-range, or even multiple databases, it is expected that amber will perform multiple sql search tool calls to break down your request and combine the results to compose the final answer.

• amber is validating the generated query for safety and accuracy, so it is not uncommon that an invalid query does not pass our validation criteria. In this case, you might see that one or more following retries are triggered, based on the complexity of your request.

Simple vs complex requests

• Simple/straightforward questions will likely be completed in faster responses with fewer retries or might need less context from your end. If you don't get the expected results, consider breaking down your broad question into smaller ones or asking for a smaller time range.

• Example simple queries:

• "How many leads do we currently have in the sales process?"

• "Show me the 10 most recent sales invoices"

• "Which storage unit has the fewest supplies at the moment"

• Example complex queries:

• "Give me a full business health snapshot for the last 6 months: order intake, revenue, inventory value, stock-outs, purchase lead times, return rates, and support ticket volume by month and region"

• "Build a reconciliation report for the last 2 fiscal periods that compares journal activity, ledger postings, and account balances by account number; highlight accounts where posting totals differ across sources beyond 2%"

Chat errors

• If you experience errors in chat, for example you see the error "Our systems are currently processing a large number of requests, so this feature is not available at the moment. Please try again in 5 minutes!" or the answer includes "I don't have the prior tool results in the visible context" or "I don't have any prior findings in this chat", it is possible that the database schema information together with the sql search results in this conversation are too big, and the requests are not completing because of the model's token limits have been exceeded.

• In this case, contact your administrator to visit the database card in the Admin Dashboard, and consider selecting fewer tables. Temporarily, try to avoid selecting more than 20 tables or selecting tables that have more than 30 columns.

• If you have multiple databases connected, consider keeping only one active database.

• Consider breaking down your question into smaller and simpler ones and ask the agent to combine the individual results at the end with a separate message.

Answer delays

• If the chat response times are to slow, the load of the current database might be to high. Temporarily, consider using the chat toggles and enable the database only for the questions that are targeted to SQL data.

• Keep the database toggle turned off for requests that are about internal company knowledge search, web search, etc.

Answer relevance

• amber will query your database and return an answer based on this data.

• If the returned data was only partially relevant to your query, amber will try to give a partial report.

• The answer will offer next steps to refine your question based on what data is available

• If amber performs many sql search tool calls with different queries, but eventually the answer is not fulfilling your question, consider breaking down your question into smaller/simpler ones.

• Confirm that the selected tables have clear and accurate descriptions. If the answer quality issue is combined with wrong table selection (this could be visible in the statuses with the generated sql queries or it might be mentioned in the answer itself), then it might be beneficial to include more context in the table descriptions and also to write a few column descriptions, especially for columns with short and not descriptive names. Both is possible in the database card in the Admin dashboard in the overview of the existing tables. Clicking on a specific table will show the available columns and allow for column description editing.

Answer not found

• If amber is not able to find an answer, consider the following:

• It might be possible that the desired data is not in your database at the moment.

• If you are an admin, explore the tables that are currently selected. It is possible that the question is targeting an unselected table.

• Consider the kind of business knowledge that it is captured in this database, and adapt your query based on this. If you have interacted with this database via other applications, try to use those past queries as a general guide for your prompts to amber.

• If you are looking for specific information, it is recommended to include this information as accurately as possible in your prompt. amber will try to match this in the database, but it has higher chances to retrieve it if you provide the exact value you are targeting, e.g., a product name, a customer's name. For instance, if you need to know how many products are currently on the way to customers and this status is referred to as "in transit" in your system, you can use the query "Count the products that are marked as in transit today" instead of saying a general statement like "Count the products that were shipped this week".

• An answer might not be found when the generated sql queries were invalid. You might see this in the answer or in the intermediate statuses. This issue is possible when the user is looking for some specific value in the database and amber could not create the correct SQL query with this precise value. Alternatively, this happens when the question is very complex, and this leads to the SQL queries being too complex as well and the validation process is rejecting them. In the first case, as explained above, please try to make the value you are looking for as precise as possible and confirm that there are no typos. In the second case, reduce the scope of your question so the SQL query targets fewer tables and their relationships, and retry using amber with this simpler version of your original question.

Final tips

• If technical issues persist and you are an admin, consider disabling the database for your organisation via the database card in the Admin dashboard, or enabling it for certain white-listed users.

• Feel free to approach us and send an email to our support team: support@amberoad.ai