Ask TradeDeck
Ask your operation a question. In plain English.
The question that used to take three people and half a day now takes the time it takes to type it. The assistant queries your live records, answers in words, and hands back the queries it ran so you can check the working.
What you can ask
Real questions, answered against real records.
These are not canned reports with a friendly label on top. Each one is answered by querying your own data at the moment you ask.
- Which Letters of Credit expire in the next 30 days with tonnage still undelivered?
- What did we pay in demurrage last quarter, by vessel?
- Show shipments where the delivered quantity differs from the LC by more than two percent.
- How much quota tonnage is left, and when does each window close?
- Total duties paid per bank this year.
- Which shipments arriving this month are missing documents?
How it works
It does not look up an answer. It works one out.
- 01
It reads your question
Typed the way you would say it out loud. No filters to set, no report to choose, no field names to know.
- 02
It writes a query
The assistant knows the shape of your records: credits, shipments, Bills of Lading, invoices, declarations, quotas, alerts and rates. It writes the query that answers the question you actually asked.
- 03
It reads the result and keeps going
If one query is not enough, it runs another. A question that spans credits and shipments and duties is answered across all three rather than handed back as three separate reports.
- 04
It answers in plain language
A sentence you can forward, with the figures in it. Alongside it, every query it ran, so a number that looks surprising can be checked rather than trusted.
Guardrails
It can read your data. It cannot touch it.
Giving a model access to a live database is a reasonable thing to be nervous about. Every limit below is enforced in code before a query is allowed to run, not promised in a policy document.
It can only read.
The assistant is given a single read-only tool. Anything that writes, changes or removes data is blocked before a query is allowed to run.
One question, one statement.
Chained statements are rejected outright, so a question can never turn into a sequence of instructions against your database.
It cannot run away with your database.
Results are capped and every query carries a hard timeout, so a broad question costs you a moment rather than a stalled system.
It shows its working.
Every query behind an answer is handed back with it. If a number looks wrong, you can read exactly how it was arrived at rather than taking it on faith.
It answers from your data, not from memory.
The assistant is instructed to base every figure on what a query actually returned, and to say so plainly when a question has no matching records.
What it will never do
- Change, add or delete a single record
- Run more than one statement per query
- Return an unbounded result set
- Sit on your database when a question is too broad
- State a figure that did not come from a query
Next step
Bring the question your team asks every Monday.
The one that takes somebody half a day to answer. We will ask it live, on your own records, and show you the queries behind the answer.