An engine born for Arabic, deciding within your boundaries
The NOVA engine reads the message as written: dialect, shorthand and all: extracts the intent and entities, and decides the next step within permissions you drew. Anything beyond the boundary escalates to human approval, and every step lands in the audit trail.
No action executes without its path
Every decision the engine proposes travels the same pipeline: draft, then permission check, then human review where needed: then a recorded execution.
The same pipeline applies to every agent in every flow: no exceptions, no side doors.
Reads the message as written
Most engines think in English and translate: meaning gets lost on the way. The NOVA engine starts from the Arabic itself.
Najdi, Hijazi, Gulf: and business fus'ha, with high accuracy. Your customers and employees never have to write the way "the system understands".
Order numbers, amounts, dates, names: even from unformatted text, so a message turns directly into fields a workflow can act on.
Permission before intelligence
Intelligence without explicit boundaries is operational risk. In NOVA, every agent's permissions are drawn before it acts: at the action and data level.
What an agent executes on its own, and what it escalates to human approval: you define both per system and per action, not as one platform-wide setting.
An action outside its permission never auto-executes: it escalates for approval with full context, and the decision: approval or rejection: is recorded in the audit trail.
| Support agent | Ops agent | Finance agent | |
|---|---|---|---|
| Order system | Allowed | Allowed | Denied |
| Customer records | Allowed | Approval | Denied |
| Invoices & payments | Denied | Approval | Allowed |
| Knowledge base | Allowed | Allowed | Allowed |
Answers from your knowledge, not thin air
The engine's answers are grounded in the knowledge sources your organization connects to the workspace: and each agent's access to each collection is explicit and visible.
Your policies, your guides, your systems: the engine grounds itself in what the workspace authorizes, so you always know where an answer came from.
Whatever an agent has not been granted simply never appears in its answers: a design rule that supports data minimization and audit readiness.
How the engine decides: in four steps
From incoming message to recorded trail: the same four steps, every time.
Reads the message as written
Before any decision, the engine turns free text: dialect, shorthand and all: into structured understanding: what the sender wants, and what data the message carries.
- Detects and normalizes the dialect before analysis
- Extracts the intent: what does the sender actually want?
- Captures the entities: numbers, amounts, dates, names
Picks the next step inside the boundaries
The decision is never open-ended: the engine matches the proposed action against the agent's drawn permissions: within the boundary it proceeds, beyond it it stops at a human.
- Matches the proposed action against the agent's permissions
- Within the boundary it proceeds; beyond it, it escalates for approval
- The full decision context is presented to the human reviewer
Works in your systems through connectors
Execution happens inside your actual systems: the order system, customer records, email: within each integration's scoped access.
- Executes through integrations within scoped access
- Separate development and production environments for testing before launch
- A failed step halts the path: no blind execution
Writes everything to the audit trail
Every understanding, decision and execution leaves a reviewable trace: you don't have to trust what happened, you can read it.
- Who decided, what was executed, when, and why
- Human approvals and rejections are part of the record
- Review the trail any time: before the audit and after it
The questions we hear in every demo
No: and that is not its purpose. The engine is designed to take over the repetitive steps: reading messages, extracting data, executing routine actions within explicit permissions. Sensitive decisions stay with your team through human approvals: so your people focus on what genuinely needs their judgment, and the organization sees everything that happened in the audit trail.
Najdi, Hijazi, and Gulf dialects, alongside business fus'ha: with high accuracy. It also extracts entities in Arabic: order numbers, amounts, dates, names: even from unformatted text, so your customers never have to write the way "the system understands".
It never executes automatically. An action beyond the drawn boundaries escalates to human approval with its full context: what was requested, why the agent proposed it, and what it would affect. The whole path: approval or rejection: is recorded in the audit trail.
From the knowledge sources and systems your organization connects to the workspace: your policies, guides, and data: within explicit access permissions for every agent. Whatever an agent has not been granted simply never appears in its answers; a design rule that supports data minimization and audit readiness.
The engine is part of a system
Understanding alone isn't enough: see where flows are built, how they're operated, and where the data lives.
The Canvas: visual builder
Where the flows the engine works inside are built: drag and drop, Arabic-first.
Explore the CanvasGoverned workflows
How the engine's decisions become observed operations: approvals, monitoring, and staged rollout.
See operationsDeployment & sovereignty
In-Kingdom cloud, your private cloud (VPC), or isolated on-premises: the engine runs where you decide.
Deployment optionsInternal knowledge assistant
The same engine answers your employees from your policies and guides: within explicit access permissions.
Read the solutionCustomer support automation
From a customer's message in their dialect to a recorded action in your systems: no waiting.
Read the solutionTrust Center
Our security model, our position on the Saudi data protection law (PDPL), and how we handle reviews.
Explore the governance centerTry the engine on your own messages.
Book a live demo where we run messages from your real work: and you watch the engine understand, decide within boundaries, and record every step.