Data sovereignty
Your data, your cloud, your country
IRAME runs the application, the data and the AI models where you decide: our cloud, your own cloud tenancy in the region you choose, or on-premise with on-premise models. Your data never trains a model.
IRAME's cloud
A single-tenant environment hosted by IRAME. Fastest to start.
Your own cloud tenancy
A dedicated private instance inside your Azure, AWS or other public cloud subscription. Your data never leaves it.
On-premise
IRAME runs on your own servers with on-premise AI models. No data leaves your building.
A deployment in your own tenancy and region: the model we are building for a Gulf national airline group whose data must stay in its home country.
| Item | Where it lives |
|---|---|
| Your ERP and source data | Read in place through read-only connections, or uploaded into your own environment. |
| Working data and results | Stored in the database inside your tenancy and region. |
| AI model calls | Sent to a model endpoint deployed in the same region, under terms that exclude training. |
| Documents and evidence | Kept in storage inside your tenancy, under your retention rules. |
| User access | Through your identity setup, and through your own virtual desktop (VDI) if your policy requires it. |
| Logs | An audit log of every action, kept in your environment and readable by your auditors. |
Answers to the questions legal and risk teams ask before approving an AI tool.
Your data never trains a model
IRA, IRAME's agentic AI, uses enterprise model endpoints such as Azure OpenAI, under terms that exclude customer data from training. Prompts and results never improve anyone's model.
Models run where your data runs
In the cloud, the model endpoint sits in your data's region. On-premise, IRAME runs on-premise models on your servers.
No open-web access
IRA works only on the data you connect. It does not search the open internet or third-party databases while testing your controls.
Deterministic tests, AI at the edges
A control runs as saved, versioned logic: same data, same control, same result every run. The model helps build the test and write it up; it does not decide the result.
A human approves every action
IRA cannot start a workflow, send a message, post an entry, approve or pay on its own. A person approves each control before it is saved and reviews each finding before it is reported.
Every step is explainable
Each result shows the data used, the steps taken and why a row was flagged. IRA's instructions and memory are visible in the Knowledge Hub.
01
Choose where it runs
IRAME's cloud, a dedicated private instance in your Azure, AWS or other cloud subscription, or your own servers on-premise.
02
Pin the region or keep it in-house
In the cloud, the application, database, storage and AI model endpoint are deployed in one region. On-premise, IRAME runs on-premise models on your servers.
03
Connect the data
Read-only connections to the ERP and data lake, or secure intake through sFTP, a parser or a controlled upload.
04
Set access and retention
Roles for each user, access through your VDI if required, and a retention period you set.
05
Review and audit
Your security team reviews the architecture, the SOC 2 Type II report and the penetration test results under NDA.
Audited
SOC 2 Type II
An independent CPA firm examined IRAME's controls over a full observation period against all five AICPA trust services criteria: security, availability, processing integrity, confidentiality and privacy.
Certified
ISO/IEC 27001:2022
IRAME's information security management system is certified to ISO/IEC 27001:2022 by an accredited body.
Tested
Independent VAPT
Independent vulnerability assessment and penetration testing against OWASP Top 10 and NIST guidance. Every finding tracked to closure.
Reports are shared under NDA. See security and deployment for the full list of practices.
Can IRAME run entirely inside our country?
IRAME can deploy the application, the database, document storage and the AI model endpoint in one cloud region, so data and processing stay in-country. Regions in the US, the EU, the Middle East and India are supported.
Can IRAME run fully on-premise?
IRAME can run on your own servers with on-premise AI models, so no data or model call leaves your environment.
Does any of our data leave our cloud tenancy?
When IRAME is deployed in your own tenancy, your data stays there. The AI model endpoint is deployed in the same region and called under terms that exclude customer data from training.
Which AI models does IRAME use?
IRAME uses enterprise model endpoints such as Azure OpenAI, deployed in your region. Controls run as saved, deterministic logic; the model helps build tests and write findings, it does not decide results.
Does IRAME need internet access from our environment?
IRAME needs to reach its in-region model endpoint and, for updates, its release channel. It does not search the open web or third-party databases while working on your data.
Can our people access IRAME only through our VDI?
IRAME works through your virtual desktop setup, so access and uploads can be limited to your controlled environment.
Bring your security and architecture team.
A working session on the deployment model, the data flows and the AI controls, with the reports your team needs under NDA.