Arabic-first AI agents
An AI agent that knows your business and speaks your customers’ Arabic.
Gental agents answer from your own documents and website, show the source behind every answer, and step aside for your team when they don’t know. Publish on your website today, and on WhatsApp soon.
- Early access
- Hosted in Saudi Arabia
- Arabic and English
Example conversation (transcript)
Visitor: Can I return an order after two weeks? And how long does the refund take?
Agent: Yes. You can return an order within 14 days of delivery, as long as the item is unused and in its original packaging. Refunds go back to your original payment method within 5 to 7 business days after we inspect the return.
Visitor: Can you change the delivery address on an order that has already shipped?
Agent: That isn’t covered in the documents I have, so I won’t guess. Someone from the team can help. Share your order number and the team will follow up.
Why most chatbots disappoint
Customers forgive a slow answer. Not a wrong one.
Generic bots guess, stumble over Arabic, trap people in loops and bill in ways nobody can plan for. Gental is built around the opposite habits.
The usual problem: Confident answers that are simply wrong
With Gental: Grounded, or honest that it doesn’t know
Answers come only from the knowledge you add, with a citation for each claim. When nothing relevant is found, the agent says so instead of improvising.
The usual problem: Arabic treated as a translation
With Gental: Arabic-native from the first line
Replies in the visitor’s language, handles dialects and Arabic-English mixing, renders right-to-left properly and searches Arabic text the way people actually spell it.
The usual problem: No way out to a person
With Gental: Knows when to hand over
Questions outside your documents, and the escalation triggers you define, end with a clear suggestion to reach your team.
The usual problem: Costs you can’t predict
With Gental: Usage you can see, per agent
Every model call is recorded with the provider’s own token counts and priced at list price, by agent and by task.
How it works
From your documents to a live agent in four calm steps
No code, no prompt engineering. Each step has sensible defaults you can refine.
Step 1
Add your knowledge
Upload files or point Gental at your website. Everything lives in one library, processed once and shared by as many agents as you like.
- PDF, Word, Markdown and text files up to 25 MB
- Website crawl or sitemap, up to 500 pages, re-synced daily or weekly
- One shared library: attach a source to several agents
Step 2
Give it a persona
Start from a template and make it yours: mission, tone, do’s and don’ts, step-by-step playbooks and escalation triggers. Or describe the role and let the AI persona builder draft it.
- Templates for customer service, sales, support, receptionist, HR, e-commerce and more
- Safety profiles for legal, medical and financial topics add rules that can’t be switched off
- Every persona in Arabic and English
Step 3
Test it before customers do
Write test scenarios, or start from the template’s, and run them against your agent. Each answer is checked and graded, so you see a pass rate and what to fix.
- Scenarios for answers, handoffs, off-topic and safety questions
- Automatic checks plus an AI reviewer with suggestions
- A playground that always runs your latest draft
Step 4
Publish where your customers are
Publishing freezes a version, so you can keep editing the draft safely. Add one script tag to your website and the agent appears in the right corner for your page’s language.
- Restrict the widget to your own domains
- Right-to-left and left-to-right handled automatically
- WhatsApp is coming soon
Website widget snippet <script async src="https://cdn.gental.ai/w.js" data-agent="pk_live_…" data-locale="en"></script>
Product
Everything an agent needs to be trusted
Thoughtful defaults for accuracy, language and control, all visible in one dashboard.
Grounded answers with citations
Hybrid search finds the right passages; every answer links back to the file page or web page it came from.
Says “I don’t know”, then hands over
Below a relevance threshold the agent doesn’t call the model at all. It replies with your fallback sentence and points to your team.
Arabic-native, end to end
Dialects, Arabic-English mixing, right-to-left layout and Arabic search normalization, so different spellings still match.
One knowledge library
Sources are ingested once and attached to any agent. Updates reach every agent that uses them.
Personas and playbooks
Reusable roles with tone, rules, playbooks and examples. Change one persona, update every agent at the next publish.
Exact usage and cost, per agent and task
Token counts come from the AI provider’s own reports, never estimates, and are priced per agent and per task: answers, search and knowledge processing.
Persona tests and an AI builder
Graded test runs with pass rates, and an AI assistant that drafts a persona and its test scenarios from a short description.
Hosted in Saudi Arabia, isolated by design
Your workspace data is stored in Google Cloud’s Dammam region (me-central2). Database-level isolation keeps every workspace separate.
Use cases
One builder, many front desks
Pick a starting template, add your own documents, and the agent takes it from there. These sample exchanges show the tone, not real customers.
Answer order, shipping and return questions around the clock
Grounded on your policies and product pages, the agent handles the repetitive questions and hands disputes to your team.
Does this size run true, or should I order one up?
According to the size guide the cut is slightly loose, so your usual size should fit. If it doesn’t, you can exchange it within 14 days.
Help patients with services, hours and booking, safely
The medical safety profile keeps the agent away from diagnosis and medication advice, and points to emergency services when symptoms sound urgent.
I’ve had a toothache for two days. What should I do?
I can’t assess symptoms, but I can help you book with a dentist; the earliest openings are on Sunday morning. If the pain is severe or there’s swelling, please contact emergency services right away.
Qualify interest and answer listing questions instantly
The agent answers from your listings and brochures, explains payment plans as written, and passes serious enquiries to your sales team.
What’s the down payment on a two-bedroom in the new project?
According to the payment plan, the down payment is 10% of the price, with the rest in instalments until handover. Would you like a sales advisor to contact you with details?
Guide students and parents through admissions and fees
From admission requirements to term dates, answers come straight from your handbooks, in Arabic or English.
When is the registration deadline for the second term?
According to the academic calendar, second-term registration closes at the end of the first week of January. The required documents are listed in the admissions handbook, page 6.
Make procedures and policies easy to find
Public services and HR teams can publish an assistant that explains procedures from official documents and refers personal cases to staff.
How many days of annual leave do I get after two years?
According to the HR policy manual, employees are entitled to 21 days of annual leave. For your current balance, the HR team can confirm it for you.
Security & trust
Your knowledge stays yours
Gental is built for businesses that answer to customers and regulators. Here is exactly what we do today, without borrowed badges.
Workspace isolation
Every record carries its workspace and is protected by row-level security in the database. A request for another workspace’s data finds nothing.
Data stored in Saudi Arabia
Workspace data, documents and search indexes are stored in Google Cloud’s me-central2 region in Dammam.
Aligned with the PDPL
Gental is designed around Saudi Arabia’s Personal Data Protection Law: data kept in the Kingdom, collected only as needed and separated per workspace.
Not used for training
Gental doesn’t train models on your content. Answers are generated by leading AI models through commercial APIs that exclude customer data from training.
Guarded instructions
Platform rules always come first: agents ignore instructions hidden inside documents, never reveal their prompt and stay within your business scope.
Your domains only
Limit the widget to your own websites, and every visitor session uses a short-lived token that’s separate from dashboard sign-in.
To write an answer, the passages relevant to a question are sent to the AI model provider for processing.
Pricing
Simple, usage-based and transparent
Pay for what your agents actually do. Plans are being finalized with our early-access customers; start free today.
Early accessFree
Free to start
To try Gental
- One agent
- A monthly message allowance to try it
- Website widget with Gental branding
Starter
Coming soon
For a first agent in production
- More agents and messages
- Larger knowledge library
- Remove Gental branding
- For growing teams
Growth
Coming soon
For teams with several agents
- More agents, messages and seats
- More knowledge and website pages
- WhatsApp when it launches
Business
Talk to us
For larger volumes
- Custom limits and seats
- Onboarding help
- Security review support
One clear unit: the message
Plans are measured in messages. Premium AI models will show as a visible multiplier, and WhatsApp fees from Meta will be passed through at cost.
What a plan sets
- Number of agents
- Messages per month
- Knowledge size
- Channels
- Team seats
Prices aren’t published yet. Nothing on this page is a binding offer.
FAQ
Questions, answered plainly
What exactly does a Gental agent answer from?
Only from the knowledge you attach to it: uploaded files and website pages. Each answer cites its sources. If nothing relevant is found, the agent replies with your “I don’t know” sentence instead of guessing.
How well does it handle Arabic?
Gental is built Arabic-first. Agents reply in the visitor’s language, understand dialects and Arabic-English mixing, and search Arabic text with normalization so different spellings of the same word still match. The dashboard and widget are fully right-to-left.
What happens when the agent doesn’t know?
It says so plainly and suggests contacting your team. You can also define escalation triggers in a persona, such as billing disputes, that always end with a handoff suggestion. A shared inbox for live handoff is on our roadmap.
Where is my data stored?
Workspace data, documents and search indexes are stored in Saudi Arabia, in Google Cloud’s me-central2 region. To write an answer, the relevant passages are sent to the AI model provider for processing under terms that exclude training.
Which channels are supported?
Today: a website widget you add with one script tag. WhatsApp is coming soon, and more channels will follow.
Can I test an agent before it goes live?
Yes. The playground always runs your latest draft, and persona tests grade a set of scenarios automatically, so you can see a pass rate and suggestions before you publish.
How is pricing calculated?
Pricing will be usage-based, with the message as the main unit. Paid plans are still being finalized during early access. You can start on the free plan now, and the dashboard already shows exact usage per agent.
Do I need a developer?
No. Building, testing and publishing happen in the dashboard. Adding the widget to your site is a single line of HTML that you, or whoever manages your site, can paste in.
Meet your gentle agent
Create your first agent on your own documents in an afternoon. Early access is open.