Arabic AI chatbot for Saudi organisations
Elbi reads Saudi dialect as well as formal Arabic, answers from your organisation's own documents, and replies in a right-to-left interface that genuinely mirrors, in text or by voice.
Most assistants sold into this region are English systems with Arabic bolted on, and customers can usually tell within three messages. Elbi is an Arabic AI chatbot built the other way round: Arabic is handled as its own language, and Arabic content is authored natively rather than machine-translated from English.
Why Arabic has to be handled natively
Arabic is diglossic. The formal written language and the language people actually type are meaningfully different, and Gulf, Najdi and Hijazi usage, along with the Egyptian and Levantine Arabic of a large expatriate workforce, diverge further. A customer typing in Saudi dialect is not writing Modern Standard Arabic, and a system tuned only to formal text will miss them.
That is a matching problem, and it has to be solved on the Arabic side directly rather than by translating into English and back. Code-switching makes it harder: Saudi customers routinely drop English product names, technical terms and numbers into Arabic sentences, and a system that must pick one language per message mangles every sentence that contains both.
The interface has to follow. Arabic is written right to left, which changes far more than text alignment, and a native reader notices within seconds when it has been treated as an afterthought.
What Elbi does in Arabic
Dialect and formal Arabic
Modern Standard Arabic and English natively, plus common Saudi dialect phrasing. Matching is built around how people ask, informally and often misspelt, not how documents are written.
Mixed Arabic and English
A question that carries an English product name inside an Arabic sentence is read as one question, not forced into one language.
Replies in the customer's language
Script is decisive: an Arabic message gets an Arabic answer, even mid-conversation. One-word replies like "ok" are deliberately not treated as a switch.
A widget that truly mirrors
In Arabic the launcher moves corner, the panel opens from the other side, bubbles swap, arrows flip, typography adjusts and the question mark sits where it belongs. See total customisation.
Arabic voice
Speech recognition treats dialect as the requirement, and synthesis understands the sentence well enough to supply the short vowels written Arabic leaves out. See voice-to-voice.
Arabic documents and scans
Customers can upload PDFs, scans, spreadsheets and photographs mid-conversation. Arabic OCR reads them, including pages that mix Arabic and English. See file understanding.
Separate Arabic and English personas
Tone, formality, length and what it must never say are set per agent and written separately for each language rather than translated from one.
Handover in Arabic
Asking for a person is treated as an instruction. Where the queue is configured for it, an Arabic conversation reaches an Arabic-speaking agent with the full transcript. See human handover.
How it answers from your own documents
- The language is settled firstArabic characters mean an Arabic conversation. A single short word is not enough to switch, because a mis-switch mid-conversation is jarring.
- The question is matched against your materialYour policies, product pages, price lists and procedures. An Arabic question can find material you only hold in English, because parallel content is rare in practice.
- Out-of-scope questions are declinedA relevance check runs before an answer is generated. Where your material is silent, the reply says so rather than improvising.
- The answer is drafted from what was foundIn the language of the question and the tone you configured, and traceable to the passage it came from.
- The reply is checked before it is sentFor claims the retrieved material does not support, and for personal data that should not be repeated back.
- A person is one sentence awayThe conversation moves to a live agent with its history intact.
Your material is used to answer the question that came with it and nothing else. It is not used to train the model. For the retrieval step in detail, see RAG retrieval.
PDPL and where Arabic conversations are processed
Elbi is built around the Saudi Personal Data Protection Law (PDPL), supervised by SDAIA. When Elbi runs on your site you are the controller of those conversations, and Elbetron processes them on your behalf under a written data processing agreement. In practice:
- Processing stays inside the Kingdom. The language model, speech recognition, speech synthesis and document reading run there, with no external AI provider in the path, so there is no cross-border AI processing to justify.
- Message content is destroyed within 24 hours, enforced by an hourly job with a backstop that runs even if everything else fails.
- Identifiers are redacted before storage and before the model sees the text: national ID, card and phone numbers and email addresses, in Saudi formats and both numeral systems.
- Consent is enforced by the server before chat, menu selections or file uploads are processed.
- Rights are buttons, not emails. Customers can download their data, delete it and withdraw consent from inside the widget.
The narrow exceptions, and what Elbi does not yet hold, are published in full on the Security & Compliance page.
Arabic chatbot use cases by industry
How to test any Arabic chatbot
Ask in dialect
Not formal Arabic. Type the way your customers type, including a sentence that mixes Arabic and English.
Ask something you have not told it
Adjacent to your material but not in it. A good system declines; a weak one improvises fluently.
Look at the Arabic interface
The mirrored layout, typography and punctuation, not translated strings in a left-to-right frame.
Ask where processing happens
Where the model runs, who else processes the conversation, and how long it is kept. Get it in writing.
Where Elbi is not the answer
Elbi runs on your website or your internal systems. It is not an omnichannel messaging product, and it is sized for departmental workloads rather than as a replacement for a national call centre. The comparison with other AI assistants sets out these limits in full, and the platform overview shows everything it does.
What it costs
Elbi is priced as a one-time setup plus monthly maintenance, published in full. A single-department website assistant in Arabic and English is SAR 50,000–75,000 to set up and from SAR 10,000 a month, excluding VAT. The pricing page covers all four tiers and first-year totals.
Common questions
Both. Elbi handles Modern Standard Arabic and English natively, plus common Saudi dialect phrasing. Customers do not type formal Arabic, so matching is built around how people actually phrase things rather than how documents are written.
The reply follows the most recent message. Script is decisive, so an Arabic message gets an Arabic answer even if the conversation began in English. Very short replies such as ok or yes do not switch the language, and sentences that mix Arabic with English product names are handled as one question.
From your own approved material. A relevance check runs before an answer is generated, so a question your material does not cover is declined and a person is offered instead. An Arabic question can still find material you only have in English.
Inside the Kingdom. The language model, speech recognition, speech synthesis and document reading run there, with no external AI provider in the path. Message content is destroyed within 24 hours, and customers can download or delete their data and withdraw consent from inside the widget.
Yes. Voice-to-voice recognises dialect as a requirement rather than a bonus, replies in a voice you choose per agent, lets the caller interrupt, and confirms the request back where it carries consequence. Voice is part of the enterprise tier and is otherwise quoted per case.
Keep reading
Test it in the Arabic your customers use
Thirty minutes on a slice of your real content. Ask in dialect, mix in English, and ask it things it should not answer.