NgageChat can only do what you can.
Every request runs under your own ERP login and permissions.
Whose permissions it uses
Yours, on every read and write. When the ERP refuses, you see the refusal and its reason.
Signed in as you
Never a shared service account.
The ERP decides
IBM Maximo, IFS and SAP apply their own permissions and segregation of duties.
Refusals are reported
You see the refusal and its reason.
In IFS, OpenID Connect
Authorisation code flow with PKCE, on the person’s own token.
Nothing installed in IFS
Official interfaces only, never the database. No custom entities, no core change.
Show me the payroll costs on project P-2041.
IFSPayrollCostSetSE_UNAUTHORIZED
IFS refused this read. Your permission set does not grant it, and the IFS screen would not show it either.
Which writes wait for your approval
You decide. A waiting write shows every field, and where each value came from, before it goes.
- Change any value in the conversation instead of starting again.
- Reads never wait. Only changes can.
Raise a fault report for pump P-114. The seal is leaking.
| Field | Value | From |
|---|---|---|
| Object | P-114 | Your request |
| Fault code | Leakage | IFS list of values |
| Site | 100 | IFS default |
| Priority | 2 | IFS default |
| Reported by | You | Your sign-in |
What the audit trail records
Every turn of every conversation, including the refused and failed ones.
| What is kept | Fields |
|---|---|
| Who and when | conversation_id, message_index, role, created_at |
| What was said | content |
| What ran, what came back | tool_invocations |
| Which model answered | provider, model |
| What went wrong | has_error, error_summary |
| Time and tokens | duration_ms, token_usage |
Your ERP keeps its own audit, in the person’s name.
Read what an audit trail should holdWhere the language model runs
Enterprise deployments run on Google Cloud Vertex AI.
- Zero data retention
- Caching and request logging off
- No training on your data
- Google does not train on it without permission
- Your region
- The Google Cloud region you choose
- Model choice
- Built to work with more than one model
Six questions to ask any AI vendor
Answered here for NgageChat.
Do you use a shared service account?
No. Every request runs under the signed-in user’s own ERP permissions.
Where does the assistant itself run?
In your own cloud, so ERP traffic stays inside your boundary, or run by EX10. Tell us your residency needs at the start: they shape the deployment.
Can a prompt talk it past a permission?
No. The ERP checks permissions on every call. An instruction in a prompt cannot change what the ERP allows.
Is our data used to train models?
Not on Vertex AI: Google does not train on your data without permission. For your deployment we name the provider, the model and the region in writing.
What happens when someone leaves?
Disable their ERP account. Without a valid ERP identity, NgageChat cannot reach anything on their behalf.
Which certifications apply?
Google Cloud’s certifications cover Google’s infrastructure. They do not certify Ngage or your deployment, and we do not present them as ours.
Send us your security questionnaire
We answer in writing, including what is not in place yet.