Working with your data in AI Chat

AI Chat does not only build applications — it works with the records inside them. It adds records, rewrites them in bulk, cleans up duplicates, finds what you are looking for and answers questions about your data, all from a description in plain language.

This article covers the records. For building tables, fields, formulas and automations see Building your app with AI Chat; for what AI Chat is and what it knows about your application see AI Chat.

Add records

One record or a whole batch, in a single request.

Add a contact: Jana Nováková, Acme Ltd, jana@acme.com, phone 601 234 567.

Here are 20 orders from our supplier — import them into Orders.

A batch is created in one bulk operation, so loading a list you paste in is a single step, not twenty.

Attachments, checklists and links. AI Chat can upload files and pictures to a record or clear them, fill in checklist items and tick them off, and attach or detach linked records.

Update records

Change the customer on all orders from Alfa to Alfa Ltd.

Say what stays the same. When you update several existing records, mention only the fields that should change and add that the rest stays as it is — "only change the status, leave everything else as is". A field left out of your instructions may otherwise be cleared instead of left alone.

Create or update by key. Using the table's import key, AI Chat updates the records that already exist and creates the ones that are missing. That is what you want when you re-load a list you have loaded before — no duplicates.

Renaming a dropdown option does not rewrite your data. If a status is renamed from Open to In progress in the designer, records still holding the old value keep it. Ask AI Chat to rewrite them:

Change status Open to In progress on all records.

Delete records

How many test records from last month are there?
Delete them.

Deleting records cannot be undone. There is no recycle bin for data. Ask for the count first, check that it matches what you expect, and only then ask for the deletion. (A deleted table is different — that one can be restored; just ask.)

When something in a batch fails

Errors are reported per record. One bad row does not stop the rest — you are told which rows failed and why, so you can fix those and leave the successful ones alone.

Changes are recorded in each record's history as usual, unless you explicitly ask to skip the audit log.

Find records

Ask in words and AI Chat turns it into a proper query: full-text search across a table, diacritics ignored (so "Novak" finds "Novák"), conditions combined with AND/OR and nested groups, and filters on columns from linked tables.

Two conditions are worth knowing about by name:

duplicates — finds records whose value appears more than once. The quickest way to clean up a table.
mine — finds the records belonging to the person asking, without naming yourself.

Which customers have a duplicate email address?
Show my open tasks due this week.

Ask questions about your data

AI Chat answers questions about the data rather than just fetching rows: counts, totals, averages, minimums and maximums over a column, results grouped by a field or by the year, month, week or day of a date field, and columns joined in from linked tables.

Show me total revenue by month for paid orders.
What is the average order value per customer this year?

In a grouped report the conditions apply to the main table, not to the joined ones.

Check your limits and usage

Ask how much of your plan you have consumed and AI Chat reports the current figures against your limits, per minute, hour, day and month: data transferred, response time, workflow runs and how long they took, calculated field recalculations, bulk create-or-update calls, attachment traffic, e-mails per recipient and AI tokens.

How much of my plan have I used this month?

It can also tell you, for any application, whether it came from the marketplace, which version of each installed extension you are running, whether a newer one exists, and what its release notes say.

Large data operations count towards the API limits of your plan, so a very big import may be processed in batches. If AI Chat reports Too many requests, wait a minute and try again — you have reached the request limit of your plan, not hit a bug.

AI usage itself is covered by credits — see AI Credits & Usage.

Good to know

AI Chat works under your own access. It sees and changes only the records you are allowed to see and change.

You see a preview before anything is written. AI Chat shows you what it is about to do, so you can look it over and confirm it — including on bulk updates and deletions.

Be specific. Name the exact table or records you mean, and prefer a condition you can check ("orders created before 1 June with no customer") over a vague one ("the old junk"). A vague request is more likely to touch more of your data than you intended.

A new automation does not run over records that are already there. If you have just had AI Chat build a workflow or a calculated field, existing records are filled in only when they are next saved — see Building your app with AI Chat.