Policy & knowledge
“Which approval applies to this purchase?”
Idrak retrieves the applicable policy for the user's role, presents the relevant clause and identifies what still needs a human decision.
BILYTICA / ENTERPRISE INTELLIGENCE
Idrak Enterprise intelligence within your boundary
Connect private enterprise knowledge to AI, then put governed answers, bilingual voice and agents to work. Deploy on your premises or in an isolated private cloud, with the data boundary defined for your organisation.
01
PRIVATE KNOWLEDGE
Documents · data · systems
02
IDRAK + GOVERNANCE
Access · model boundary · audit trail
03
UNDERSTAND · SPEAK · ACT
Answers and agents with human approval
FROM SOURCE TO DECISION
How private data becomes governed action
Idrak starts with your sources and permissions, not a generic model. Each use case has a defined task, a review point and a traceable result.
01
Connect private knowledge
Select documents, databases and operational systems. Define what Idrak may read, who owns each source and how it is refreshed.
02
Apply intelligence and controls
Ground responses in authorised content. Configure role-based access, model boundaries, logging and approval rules for each use case.
03
Return an answer or approved action
Deliver an insight, a bilingual voice response or an agent task. Keep a record of the source, decision and approval.
YOUR INPUTS
Documents and policies
Databases and analytics
ERP and CRM systems
IdrakBoundary & access
REVIEWABLE OUTPUTS
Understand
Speak
Act
Intelligence in practice
Understand
Answer questions against approved internal knowledge, with access tied to the user's role.
Speak
Support Arabic and English voice interactions grounded in your organisation's terminology.
Act
Route tasks through connected systems, with a person approving consequential actions.
ON-PREMISES ARCHITECTURE
Your data, your model boundary, your decision
Idrak is more than a chat window on a public model. In an on-premises configuration, retrieval, model runtime and interfaces are deployed inside your defined environment. Any necessary external connection is reviewed rather than assuming every configuration is fully disconnected.
WITHIN YOUR NETWORKBOUNDARY DEFINED PER DEPLOYMENT
01
Your knowledge
Policies, contracts, manuals, databases and operational systems; each source has an owner and access policy.
02
Permission-aware retrieval
Find relevant passages from approved sources, then filter against the requesting user's access.
03
Private AI runtime
Run a right-sized model against retrieved context; model choice, hardware and outbound access are agreed per deployment.
04
Reviewable response or task
Return a referenced answer, bilingual voice reply or proposed action. Sensitive changes wait for human approval.
Identity & accessLogging & reviewHuman approval for consequential actions
Reference architecture, not a promise of ready-made integrations or automatic regulatory certification. Your technology and security teams define the actual controls.
USE CASES
A specific job, not an open-ended demo
Each use case begins with the source of truth, the decision owner and the allowed action. These illustrative workflows are not published customer results.
Operations & service
“Summarise this case in Arabic and prepare the next step.”
With permitted case records, Idrak prepares a sourced summary and proposes a task; the team reviews before any system update.
Voice at work
“Explain this procedure while I am in the field.”
A bilingual voice interface can return guidance based on the current approved procedure, with terms tested against the organisation's language.
THE DECISION PATH
An answer is not an action
The model can answer from authorised material. Changing a record, sending a request or triggering another system follows a separately defined approval path.
Authorised request
Idrak retrieves accessible content and records its source.
Sourced answer
Return the answer, or prepare an action if the use case permits it.
Approval before change
An authorised person reviews consequential changes; the decision and result are recorded.
FROM DESIGN TO OPERATIONS
What makes on-premises AI workable?
Connections are scoped
Start with an inventory of systems, available interfaces and data classification. Connect only approved sources; do not assume that every ERP or database has a ready-made connector.
Access follows the source
Map identities and roles before indexing. Retrieval and agent tools are scoped to the person's entitlements; test access with real user roles before wider use.
Actions have a checkpoint
Define which tasks are read-only and which can write back. For consequential updates, show the intended change, ask an authorised person to approve it and record the outcome.
Operations are owned
Agree who refreshes source indexes, updates models, reviews logs and handles failures. Offline environments need a defined route for patches and approved content updates.
Sizing also considers model compute, storage, backup, capacity and service continuity. These requirements follow the real workload rather than a one-size-fits-all appliance.
DEPLOYMENT DECISION
Define the data boundary before deployment
On your premises
Run models, knowledge and interfaces inside your own network. An air-gapped configuration can be evaluated when your data classification calls for it, including how updates and support will work.
Isolated private cloud
Deploy in a dedicated cloud environment with agreed network, access and logging controls. Data location and connectivity follow your classification and organisational requirements.
PDPL, NDMO and cybersecurity controls are assessed against the data, sector and deployment. A product label is not a substitute for a compliance review.
Idrak questions, answered
What is Idrak?
Idrak is Bilytica's sovereign enterprise intelligence platform. It connects private organisational knowledge to governed answers, bilingual voice interactions and AI agents that can work with internal systems.
Does Idrak send our data to a public AI service?
An on-premises or isolated private-cloud configuration can be designed without external model calls. The network boundary, integrations and any exception are agreed and tested for each deployment; not every configuration has the same boundary.
Which systems can Idrak connect to?
The reference architecture covers documents, databases, analytics dashboards, ERP and CRM. Available connectors and permissions are confirmed during discovery, rather than assumed for every customer.
Can Idrak work in Arabic and English?
The platform is designed for Arabic and English text and voice. We test institutional terminology and Saudi-language examples against the customer's own material before use.
Is Idrak compliant with PDPL and cybersecurity controls?
Deployment can support a customer's PDPL, NDMO and NCA control requirements through data classification, access rules, network boundaries and audit evidence. Compliance depends on the organisation's configuration and independent review, not a product label.
Start with a use case from your data
Discuss your knowledge sources, deployment boundary and agent permissions with our team.