AWRA OpsHub Search
AI data controls

Switch third-party AI off for your organization. See every call it made.

One switch under Settings › AI & Data stops every in-app feature from sending anything to an AI provider. Ten module switches turn AI off where you want it quiet. And a usage log shows the last 30 days of calls by provider and purpose, so you can see where your data went without anyone keeping a copy of what was in it.

Workspace-wide switch · Ten module switches · Providers named · 30-day usage log · Content never logged · Web and mobile app

Settings › AI & Datayour organization
On
  • Insights, forecasts and quotation comparisons add a written summary.
  • The helpdesk copilot drafts, summarises and triages.
  • Each answered call adds a row: provider and purpose.
Off
  • Nothing is sent to an AI provider, from any in-app feature.
  • Insights, forecasts and comparisons still show their figures.
  • The change itself is written to the audit log, with who made it.

Changing it needs the edit-settings permission. Everyone in the organization is affected at once.

One doorevery language-model call in the product passes one place, and that is where the switch is checked
Two providersGroq first, Google Gemini when Groq is unavailable, called directly with no aggregator in between
Provider and purposelogged for each answered call, with the prompt’s length as a number, never its text
Figures stayforecasts and quotation scoring run on AWRA’s own service and keep working with the switch off
Two positions

What the product does with the switch on, and with it off.

Switching third-party AI off removes the written layer, not the work underneath it. The figures are computed by AWRA either way; what a provider adds is the sentence that explains them, the draft that saves typing, or the rephrased answer.

Feature
Third-party AI on
Third-party AI off
Dashboard and module insights
OnThe computed figures and alerts, plus a short written summary naming the one action to take next.
OffThe same figures and alerts, without the written summary.
Demand forecasts
OnForecasts from AWRA’s own service, with a narrative explaining them.
OffThe same forecasts from AWRA’s own service, without the narrative. Same figures
Quotation comparison
OnQuotes scored on AWRA’s own service, with a written analysis of the result.
OffThe same scoring, without the written analysis. Same ranking
In-app assistant guidance
OnThe matched in-app guidance, rephrased by a model that is given only that guidance to work from.
OffThe matched guidance as it is written, with the same links to the screens it names.
Questions about your own figures
OnApprovals, low stock, overdue invoices and the like, answered from your own records. These answers never pass through a model.
OffAnswered the same way, from your own records. Unchanged
Helpdesk copilot
OnReply drafts, ticket summaries and triage suggestions, each landing where an agent reviews it.
OffCopilot requests return nothing; agents read, triage and reply as they do on any helpdesk.
Supplier prequalification
OnAn AI assessment of each application, alongside the reviewer’s own.
OffApplications arrive and are reviewed, approved or declined by people, without the AI assessment.
Workflow builder
OnSuggestions while you build a workflow.
OffThe builder without suggestions; every workflow you have keeps running.
Usage log
OnA row for each answered call: provider, purpose, and the prompt’s length as a number.
OffNo new rows, because no call is made. The last 30 days stay readable.
The control panel

The switch, the modules, the providers and the log, on two screens.

Settings › AI & Data holds the workspace switch, the providers that receive requests and the usage log. Settings › AI modules holds a switch per module. The mobile app carries the same switches, over the same API.

Switchesillustration
Use third-party AIFor summaries, assistants and analysis, across the whole organization.
  • InventoryStock insights, demand forecasts, smart reorder, stock answers
  • ProcurementVendor insights, quotation analysis, supplier evaluation
  • SalesSales insights, overdue-invoice and sales answers
  • AccountingAccounting insights and their summaries
  • HRPeople Pulse insights, summary and HR answers
  • ProjectsProjects insights and their summary
  • AssetsAsset insights on the asset screens
  • HelpdeskReply drafts, summaries, triage, support insights
  • POSPoint-of-sale insights
  • AutomationWorkflow run health, failures and its summary
A module switched off shows its insight cards as turned off, and the assistant says where to switch it back on.
Who receives what, and whenillustration
FirstGroqLow-latency inference for the assistant and analysis features. United States.
When Groq is unavailableGoogle GeminiInsights, summaries and recommendations. Global (US).
AI calls in the last 30 days, by purpose and provider
PurposeProviderCalls, 30 daysLast callContent
In-app assistantGroq41228 Oct 2026 09:14not stored
Dashboard and module insightsGroq23628 Oct 2026 08:02not stored
Helpdesk drafts and summariesGroq8827 Oct 2026 17:45not stored
Forecast narrativesGroq4127 Oct 2026 06:30not stored
Dashboard and module insightsGoogle Gemini921 Oct 2026 11:20not stored
Quotation analysisGroq619 Oct 2026 14:08not stored

The content column is drawn here to make the point. The real screen has no such column, because there is nothing to put in it: the log holds the provider, the purpose, the time and the prompt’s length in characters.

The provider list is read from the published sub-processor register, so the settings page and the public page say the same thing.
How it works

Every model call goes through one door, and the switch is on the door.

A switch checked feature by feature is a switch some feature forgets. Here the check sits in the one service every language-model call passes through, so a feature added next month is covered the day it ships.

1

A feature asks for words

An insight wants its summary, an agent asks for a draft, the assistant has matched a piece of guidance. The figures are already computed by then.

2

The door checks your organization

Is third-party AI on, and is there room in the plan’s monthly AI allowance? If either answer is no, nothing is sent and the feature shows what it has without the written part.

3

Groq, then Gemini

The request goes to Groq. If Groq is unavailable it goes to Google Gemini. Each is called on its own API, with no gateway in between.

4

One row in the log

When a call is answered, the log records the provider, the purpose, the time and the prompt’s length. The prompt and the answer are not written to it.

Who holds the keys

A setting for everyone, changed by the people you choose.

The workspace switch

Seen by anyone with the view-settings permission and changed only by a role with edit-settings. Administrators hold both.

The module switches

A separate permission, manage_ai_settings, because turning a module’s AI off changes it for everyone who uses that module.

Recorded when it moves

Switching third-party AI on or off writes an entry to the audit log naming the person who did it, so the question of when AI was enabled has an answer.

Same switches on mobile

The mobile app shows the switch, the module switches, the providers and the usage, under the same permissions as the web.

Counted against your plan

Each answered call counts toward the plan’s monthly AI allowance. When it is used up, features carry on without the written layer until the next month.

The public assistants are separate

The website and Academy assistants answer visitors from published material, not your organization’s records, so they sit outside this switch.

The straight answer

What AWRA OpsHub does today

  • A workspace switch for third-party AI under Settings › AI & Data, checked in the one service every language-model call passes through, so no in-app feature sends anything to a provider while it is off.
  • Features that keep their figures with the switch off — insights, forecasts and quotation comparisons show what they computed, without the written summary.
  • Forecasts and quotation scoring on AWRA’s own service, unaffected by the switch.
  • Answers about your own figures in the in-app assistant, taken from your records and never passed through a model.
  • Ten module switches — inventory, procurement, sales, accounting, HR, projects, assets, helpdesk, point of sale and automation — under their own permission.
  • The providers named on the settings page with what each is used for, its region and the data it receives, read from the published sub-processor register.
  • A 30-day usage view of calls by provider and purpose, with the count and the time of the last call; the content of a request is never logged.
  • An audit log entry each time the workspace switch is turned on or off, naming who did it.
  • The same switches and usage in the mobile app, over the same API.

More we can add to your workspace

  • A per-call view of the usage log with each call’s time and screen, and a CSV export of it for your own records.
  • A date range on the usage view reaching further back than the last 30 days.
  • A choice of provider per organization, such as Gemini only, or Groq only.
  • AI usage events in the audit log stream, so your SIEM sees each call beside the rest of the trail.
  • A switch per role, so AI summaries reach managers while other roles see figures only.
  • A self-hosted inference option, answering requests inside our own environment with no third-party provider involved.

Where we point you to a specialist

  • We do not keep the content of an AI request in the usage log. A log of what was sent would be a second copy of your records in a place nobody reviews, which is the opposite of what a control is for. The log says who received a request and why; the record itself stays where it lives.
  • We do not route requests through a model aggregator. Each provider is called on its own API, so the list on the settings page is the whole list.

The per-call view with a CSV export is the item that most often decides a security review, and the audit-stream events are the one that puts AI usage in front of the team that already watches everything else. Tell us what your reviewers ask for, and we will come back with a written spec, a timeline and a price.

Frequently asked questions

What happens when we switch third-party AI off?
Nothing is sent to an AI provider from any in-app feature. The check sits in the one service every language-model call passes through, so insights, the in-app assistant, helpdesk drafts, quotation analysis, supplier assessment, forecast narratives and workflow suggestions are all covered. Each feature shows what it computed without the written part: insight cards keep their figures, forecasts and quotation scores are unchanged, and the assistant serves its guidance as written.
Which AI providers receive requests?
Groq first, and Google Gemini when Groq is unavailable. Each is called directly on its own API, with no model aggregator in between. The settings page lists them with what each is used for, its region and the data it receives, read from the published sub-processor register.
What does the usage log record?
For each call that is answered: the provider, the purpose (for example the in-app assistant, or helpdesk drafts and summaries), the time and the length of the prompt in characters. The settings page shows the last 30 days grouped by provider and purpose, with the number of calls and the time of the last one. The text of the prompt and the answer are not stored in it.
Who can turn third-party AI on or off?
Anyone with the view-settings permission can see the setting; changing it needs edit-settings, which administrators hold. The module switches need a separate permission, manage_ai_settings. Turning the workspace switch on or off is written to the audit log with the name of the person who did it.
How are the module switches different from the workspace switch?
The workspace switch stops calls to third-party providers and leaves every feature in place with its figures. A module switch turns that module’s AI features off altogether — its insight cards show as turned off and the assistant’s answers for that module point to the setting — whether or not third-party AI is on.
Does anything still work with third-party AI off?
Yes. Demand forecasts and quotation scoring run on AWRA’s own service. Insight cards show their computed figures and alerts. The in-app assistant still matches your question to in-app guidance and links to the right screen, and answers questions about your own figures — approvals, low stock, overdue invoices — straight from your records, which it never passes through a model in any case.
Is our data used to train AI models?
We do not use your prompts, documents or records to train, fine-tune or improve any model, and we select and configure providers so that inputs are not used for their model training. The AI trust page sets out the models in use and what is sent to them, and the sub-processor page lists the providers.
Do the assistants on the public website follow this switch?
No. The website and Academy assistants answer visitors from published material rather than from your organization’s records, so they sit outside the workspace switch. Everything a signed-in person uses inside your workspace is covered by it.

Help Center

Need a quick answer while you read?

Run inventory, procurement, assets, sales, and field work with approved AWRA guidance for setup, migration, integrations, security, pricing, and support.

Search all approved AWRA public help articles.

Open Help Center