How a question reaches IFS Cloud data
From a typed sentence to an OData request, step by step.
The short answer
Find the projection, read its fields, build the query, run it as you.
How does a sentence become a query?
Each step checks IFS before the query runs.
-
Find the projection
The entity set that holds the data.
-
Read its fields
Keys, types, enumerations and actions.
-
Build the query
Filters on fields that exist.
-
Run it as you
With your own IFS access.
What is a projection in IFS?
A published API over part of the IFS data model. It is also where IFS grants access: a permission set gives each projection none, read-only, custom or full access.
Why does that matter for AI?
IFS decides and enforces access at the projection. An assistant that calls IFS as the real person inherits the right answer to who may see what, without access rules of its own.
What if it uses a service account?
Then it is exactly as capable as that account. Point it at a service account with full grants and you have built a way around your permissions, with a chat box on the front.
How do writes differ from reads?
A write needs every mandatory field, and often your approval.
| Step | For a read | For a write |
|---|---|---|
| Fields | Only those it returns | Every mandatory field, resolved |
| Values | Filters on fields that exist | Enumerations and lists of values read from IFS |
| Defaults | Not needed | IFS’s own _Default() function |
| Concurrency | Not needed | The record’s current ETag, sent with the change |
| Approval | Not needed | A summary first, where you require a yes |
Does it query live data?
Yes. The request goes to your IFS Cloud OData endpoints when you ask, so the figure is the one IFS holds then. No replica, no overnight extract.
When is live querying the wrong tool?
Live queries need a connection to IFS, and very large aggregations are bounded by what the OData layer returns efficiently. For heavy analysis across years of history, a reporting platform is still the right tool.
When is it the right one?
For the questions people ask all day, such as what is late, what is short and who owns it. Live is fast enough for those, and more useful than a copy that was true last night.
More questions from IFS teams
Does this need database access to IFS?
No. Everything goes through the published IFS OData REST APIs. Nothing is installed in IFS, there is no database connection and no change to core code, so your upgrade path stays clean.
What happens if the question is ambiguous?
It asks. If a supplier name matches three suppliers, or an asset ID matches two objects, a well built assistant shows the candidates instead of picking one. A clarifying question beats a confident wrong record.
Can it handle custom fields?
Yes. Custom fields differ by tenant, so NgageChat reads them live from your environment’s configuration APIs instead of assuming a fixed IFS data model.
Why does it matter how an action is called?
IFS has bound actions on one record, such as releasing a work order, and action imports on the projection, used by wizard style flows such as expense sheets. Calling a function the wrong way can return no records instead of an error, the kind of silent wrong answer that costs trust.
See NgageChat answer a question from your IFS
Send us a task you have not prepared for. We run it with you, live, on a test system.