AIBI System
Security and deployment

Deployment that fits your infrastructure and security requirements

SaaS, installation in your infrastructure, or a fully isolated environment with no external LLM.

Deployment

Deployment options

Choose where the system runs and whether any data reaches an external LLM.

SaaS

Quick start

AIBI cloud

  • System and database hosted by AIBI
  • Data delivered on a schedule via API
  • Minimal setup, fast to launch
Best for a pilot

On-Premise

External LLM

Your infrastructure

  • System runs in your infrastructure
  • AI requests go to an external LLM (via our API)
  • You control the data and the pipeline
Higher infrastructure requirements

Full On-Premise

Internal LLM

Fully isolated environment

  • Runs entirely in your infrastructure
  • Your own inference server, or deployment on your GPUs
  • No data sent to an external LLM
For strict security requirements
Across all optionsSecurity and data control
Data masking before the LLMPrompt injection protectionRoles and permissionsData quarantine
Data flow

How data moves through the system

Data sourcesyour infrastructure
Quality checks · quarantineAIBI cloud
StorageAIBI cloud
external LLM
AI analysis · maskingAIBI cloud
Insights · charts · alertsAIBI cloud

Storage and analysis run in the AIBI cloud. Data arrives on a schedule and is masked before it reaches the external LLM.

Scales to your data volumes. The storage architecture is chosen to fit your data and infrastructure.

Access and control

Enterprise access control

01

Roles and permissions

Control access to data, sections and features by user group.

02

SSO

Sign in through your corporate identity provider (OIDC supported).

03

Audit log

A record of user activity: data access, exports and changes.

FAQ

Security questions

In SaaS and On-Premise, sensitive data is masked before any request reaches the LLM. In Full On-Premise, data never leaves your infrastructure (the LLM and inference run internally).

Personal and commercially sensitive fields, such as names, counterparties and amounts, based on your rules. Masking happens before anything is sent to the model.

Yes, with Full On-Premise. Both the system and the model run in your infrastructure.

Through user roles and permissions that control access to data and sections.

Yes. A dedicated protection layer guards against prompt injection and blocks resource-heavy database queries generated by the LLM.

The engine runs on models we selected because they perform best on analytical tasks. If needed, we can connect a different model, including your internal one (Full On-Premise).

Start with a pilot on a limited historical slice of your data.

Let's plan a deployment for your environment.

We'll help you choose SaaS, On-Premise or Full On-Premise to meet your security team's requirements.

Discuss deployment