منصة
Platform
One platform. Every front line.
NOOR is a multi-agent operating system for customer contact: five agent types, three channels, and a rule layer that makes AI answers safe for businesses that cannot afford to hallucinate a price.
How NOOR works
A message from any channel lands in one conversation state. Your agent decides what to say — constrained by your rules. The outcome lands in your system of record. A human is one click away, always.
Channels
WhatsApp Cloud API, web widget, and telephony (KSA/UAE). One conversation graph per customer, regardless of where they start.
Agents
Each agent is a persona with approved facts, mandatory reply templates, and hard boundaries. Language lock, escalation rules, and tone per customer.
System of Record
Bookings, leads, and transcripts land in your CRM and calendar. Structured extraction: names, numbers, dates, amounts, intent — per conversation.
Autonomy
Deterministic rules first, model reasoning second, humans last. The platform enforces guardrails at the code layer — not by prompting the model to behave.
The NOOR guarantee
Guardrails enforced by the platform — not by prompt
Most AI chatbots are one bad model day away from quoting a price you never approved. NOOR inverts that: the model proposes, the rule layer disposes. Here is how, concretely.
Approved facts only
An agent can only assert what is on its facts list: your price sheet, room list, policies. Anything else gets a fixed, honest reply and goes to your team. This is enforced by the platform — not by prompt.
Mandatory replies
For sensitive outcomes — a recorded reservation, a forwarded request, an unavailable property — the exact wording you wrote is the exact wording the customer hears. No paraphrase, no drift, no surprises.
Full audit trail
Every reply is logged with the rule or agent decision that produced it. Dispute a booking? You have the transcript, the extraction, and the rule trace.
A day with NOOR
23:41 → 09:00 → 14:00
A hotel in Al Olaya. 23:41: a guest asks in Arabic (Najdi) for the Royal Suite tonight for two. NOOR replies in the same language, quotes exactly the approved price, and records the request. It does not promise the room — its scripts say what it may say.
09:00: the front desk opens. Three requests, all transcribed, all with names and numbers extracted. Two confirmed, one declined (the suite is booked). NOOR sent the declined guest the exact follow-up script: no invented alternatives, no apology theater — just a clean forward to the team.
14:00: the team checks the board: 14 conversations, 11 resolved, 3 escalated. Top question this week: “Is breakfast included?” — a one-line addition to the facts list fixes it before Monday.
See the hotel use case23:41:02 WHATSAPP inbound · from +966 5X XXX XXXX
guest: «أبغى سويت ملكي الليلة لشخصين، كم السعر؟»
23:41:04 agent=Front Desk · lang=ar · locale=ar-SA
23:41:04 RULES facts:room.royal.suite ✓ · reply_template=pricing_query
noor: «السويت الملكي ٢٬٨٠٠ ريال لليلة مع الإفطار. يتم تسجيل الطلب وفريقنا سيتواصل للتأكيد.»
23:41:05 EXTRACT { room: "Royal Suite", guests: 2, date: "today", intent: "reserve" }
23:41:05 ACTION booking_request.recorded → queue: front-desq-morning
# no availability asserted · no confirmation sent · template verbatim
نور
See the platform on your channel first.
We run a supervised shadow week before any agent goes live: NOOR answers in parallel, your team sees everything, you approve the handover.