Your data. Your AI. Your region.
RiskSafetyAI runs AI on infrastructure you own — your cloud account, your region, your monitoring, your content policy. Your safety data never touches an AI provider we control.
Your safety data never touches an AI provider we control.
The three questions we get asked first
Will our data train your AI?
No
— and structurally it can’t, because we don’t operate the model. Every AI call runs on your provider account, under the data processing agreement you already hold with Microsoft, AWS, or whoever you choose. No new vendor relationship to underwrite.
Can your AI see another mine’s data?
No
Every request is bound to your tenant at four independent layers, from token validation through to the data layer. No shared index, no pooled retrieval. On the dedicated tier your database is physically separate as well.
Where does our data physically sit?
In the region you pick
Database, files and audit log together — AWS Cape Town or Azure Johannesburg for South African sovereignty. Because you also choose where inference runs, the entire path stays in one jurisdiction.
We never own the AI relationship. You do.
Most safety platforms route your data through the vendor’s AI account and ask you to trust their policy about what happens to it.
RiskSafetyAI is built the other way around. You supply the endpoint and the credential. From that moment every AI call from your organisation runs through your Azure subscription or AWS account — billed to you, logged in your monitoring, filtered by your content policy, and switched off the moment you disable the configuration.
Your bill
token spend is a line item your FinOps team can see
Your logs
every invocation lands in your Azure Monitor or CloudTrail
Your policy
your content filters and guardrails apply to every prompt
Your switch
disable the provider and AI fails closed
Don’t take our word for it. Have your cloud team watch the logs. Configure the provider, run one AI assessment, and confirm the invocation appears in your own monitoring — and nowhere else.
Policy requires no outbound internet at all? Point us at a self-hosted model inside your network — one of nine supported AI providers. That configuration has no external AI path whatsoever.
You set the ceiling. AI spend cannot cross it.
The real question behind “what does the AI cost?” is usually “what happens when it runs away?” It can’t — and not because we promise it won’t. Four controls are built into how the platform runs.
A hard monthly cap
Set a monthly AI budget per site. When it’s reached, the platform stops making AI calls and falls back to standard rules. The cap is a ceiling, not a warning threshold.
Off unless you turn it on
Automated review is opt-in, per site and per form type. All twenty-seven AI capabilities enable and disable independently — you pay for what you switched on and nothing else.
The cheap engine does the heavy lifting
High-volume automated review runs on the most cost-efficient model tier. The more capable, more expensive tier is reserved for occasional deliberate authoring tasks.
Never in the critical path
AI runs in the background, decoupled from form submission and offline sync. It cannot slow down, block or break the work your crews depend on — and if it’s paused, nothing stops.
AI assessment costs in the order of 3 cents per form. Less than printing the paper form it replaces.
Indicative, at current provider pricing — and easy to size yourself: multiply by your own daily form volume. There is no per-user fee and no platform uplift on inference; you pay your AI provider directly, at their rate.
| Forms assessed per day | Indicative AI cost per site, per month |
|---|---|
| Around 25 — a small site or a single crew | ≈ R450 (~$23) |
| Around 100 — a typical operating site | ≈ R1 700 (~$90) |
| Around 400 — a large or high-compliance site | ≈ R6 700 (~$360) |
Indicative monthly AI cost by site throughput. Confirmed against your real volumes in discovery. Illustrative, at ±R18.5/$ — your provider bills in USD.
Around 25 forms/day
a small site or a single crew
≈ R450 (~$23)
Around 100 forms/day
a typical operating site
≈ R1 700 (~$90)
Around 400 forms/day
a large or high-compliance site
≈ R6 700 (~$360)
The relationship is linear and shallow. Ten times the forms is ten times a small number — and whatever figure you land on, your monthly cap is still the ceiling.
Expect a one-off spike in the first month while your existing procedures and documents are ingested. That happens once, then effectively stops — it’s a setup cost, not a running cost.
The cap never breaks the platform. Reach your monthly ceiling and AI simply pauses; every form, inspection, permit and report keeps working on standard rules for the rest of the month. No surprise invoice, no service interruption.
Two ways to run it. Both in your cloud account.
The AI provider account is always yours. Your subscription, your billing relationship, your region, your kill switch. What you choose is who does the work of setting it up and keeping it running.
There is no AI markup, because we never resell inference.
| Dimension | Managed by us | Managed by you |
|---|---|---|
| Where it runsIdentical either way | Your Azure or AWS tenant | Your Azure or AWS tenant |
| Who configures it | We do — provisioning, model selection, tuning, monitoring | Your cloud team |
| Who pays the AI vendorIdentical either way | You, directly at cost | You, directly at cost |
| Who holds the kill switchIdentical either way | You | You |
| Best for | Teams without spare cloud engineering capacity | Teams with an established cloud practice and their own governance |
Where it runs
Identical either wayYour Azure or AWS tenant
Your Azure or AWS tenant
Who configures it
We do — provisioning, model selection, tuning, monitoring
Your cloud team
Who pays the AI vendor
Identical either wayYou, directly at cost
You, directly at cost
Who holds the kill switch
Identical either wayYou
You
Best for
Teams without spare cloud engineering capacity
Teams with an established cloud practice and their own governance
Or point us at a self-hosted model inside your own network — no per-use AI fees at all, and no external AI path.
Where we manage the AI in your tenant, we operate under a least-privilege service principal scoped to the AI resource alone, issued by you and revocable by you at any time. We never hold standing access to anything else in your subscription.
Three ways to run it
Multi-tenant SaaS
we operate it, you consume it. Fastest to live.
Dedicated single-tenant
your own database and instance, in the AWS or Azure region you choose.
On-premise / private cloud
entirely inside your estate, on your VMs or Kubernetes.
Same codebase in all three. Same features, same audit trail, same AI options — the choice is about isolation and who operates the infrastructure, not about capability. Regional hosting addresses POPIA, GDPR, UK GDPR, the Australian Privacy Act, and UAE and KSA PDPL residency obligations.
Give this to your enterprise architect
We maintain a full architecture document — tenant isolation, identity, encryption, AI data flow, deployment models, integration surface. Ask for it and we’ll send it to your IT lead before we meet, then walk them through it in the session. If it doesn’t stand up to their reading, we’ve saved you a meeting.
Book a demo